Connexion
AMD has made itself quite a reputation with its bundling campaigns over the years, and every new season we can be sure that the company will be giving away free games with the purchase of its hardware. This summer will certainly not be exception as AMD will be bundling Warhammer 40,000: Space Marine 2 and Unknown 9: Awakening titles with its Ryzen 7000 CPUs and Radeon RX 7000 video cards.
The latest bundle offer essentially covers all of AMD's existing mid-range and high-end consumer desktop products, sans the to-be-launched Ryzen 9000 series. That includes not only AMD's desktop parts, such as the Ryzen 9 7800X3D, but also virtually their entire stack of Radeon RX 7000 video cards, right on down to the 7600 XT.
AMD's laptop hardware is also covered as well, which is a much rarer occurence. Mid-range and high-end Ryzen 7000 mobile parts are part of the game bundle, including the 7940HS and even the 7435HS. However the refreshed version of these parts, sold under the Ryzen 8000 Mobile line, are not. Meanwhile systems with a Radeon RX 7700S or 7600S mobile GPU are included as well.
This deal is available only through participating retailers (in case of the U.S. and Canada these are Amazon and Newegg). The promotion is also applicable to select laptops containing these components.
| AMD's Summer 2024 Ryzen & Radeon Game Bundle (Warhammer 40,000: Space Marine 2 & Unknown 9: Awakening) |
||
| CPU | GPU | |
| Desktop | Ryzen 9 7950X3D Ryzen 9 7950X Ryzen 9 7900X3D Ryzen 9 7900X Ryzen 9 7900* Ryzen 7 7800X3D* Ryzen 7 7700X* Ryzen 7 7700* |
Radeon RX 7900 XTX Radeon RX 7900 XT Radeon RX 7900 GRE Radeon RX 7800 XT* Radeon RX 7700 XT* Radeon RX 7600 XT* |
| Laptop | Ryzen 9 7940HS Ryzen 7 7840HS Ryzen 7 7735HS Ryzen 7 7435HS |
Radeon RX 7700S Radeon RX 7600S |
| *This product does not qualify for the promotion in Japan | ||
Warhammer 40,000: Space Marine 2 carries an MSRP of $60, whereas the Unknown 9: Awakening is set at $50, so this offer provides an estimated value of $110. The deal is particularly appealing to gamers and those interested in action titles. Meanwhile, fans of such games probably already have AMD's Ryzen 7000 and Radeon RX 7000-series products, so while the deal will be appealing to some users, it may not be appealing for gamers looking to upgrade to AMD's latest Zen 5-powered CPUs.
The campaign starts on August 6, 2024, at 9:00 AM ET and ends on October 5, 2024, at 11:59 PM ET, or when all Coupon Codes are claimed, whichever happens first. Coupon Codes must be redeemed by November 2, 2024, at 11:59 PM ET.
As sales of GPU-based AI accelerators remain as strong as ever, the immense demand for these cards has led to some server builders going off the beaten path in order to get the hardware they want at a lower price. While both NVIDIA and AMD offer official card configurations for servers, the correspondingly high price of these cards makes them a significant financial outlay that some customers either can't afford, or don't want to pay.
Instead, these groups have been turning to buying up consumer graphics cards, which although they come with additional limitations, are also a fraction of the cost of a "proper" server card. And this week, ASRock has removed another one of those limitations for would-be AMD Radeon users, with the introduction of a set of compact, passively-cooled Radeon RX 7900 XTX and RX 7900 XT video cards that are designed to go in servers.
Without any doubts, ASRock's AMD Radeon RX 7900 XTX Passive 24GB and AMD Radeon RX 7900 XT Passive 20GB AIBs are indeed graphics cards with four display outputs and based on the Navi 31 graphics processor (with 6144 and 5376 stream processors, respectively), so they can output graphics and work both with games and professional applications. And with TGPs of 355W and 315W respectively, these cards aren't underclocked in any way compared to traditional desktop cards. However, unlike a typical desktop card, the cooler on these cards is a dual-slot heatsink without any kind of fan attached, which is meant to be used with high-airflow forced-air cooling.
All-told, ASRock's passive cooler is pretty capable, as well; it's not just a simple aluminum heatsink. Beneath the fins, ASRock has gone with a vapor chamber and multiple heat pipes to distribute heat to the rest of the sink. Even with forced-air cooling in racked servers, the heatsink itself still needs to be efficient to keep a 300W+ card cool with only a dual-slot cooler – and especially so when upwards of four of these cards are installed side-by-side with each other. To make the boards even more server friendly, these cards are equipped with a 12V-2×6 power connector, a first for the Radeon RX 7900 series, simplifying installation by reducing cable clutter.
Driving the demand for these cards in particular is their memory configuration. With 24GB for the 7900 XTX and 20GB for the 7900 XT is half as much (or less) memory than can be found on AMD and NVIDIA's high-end professional and server cards, AMD is the only vendor offering consumer cards with this much memory for less than $1000. So for a memory-intensive AI inference cluster built on a low budget, the cheapest 24GB card available starts looking like a tantalizing option.
Otherwise, ASRock's Radeon RX 7900 Passive cards distinguish themselves from AMD's formal professional and server cards by what they're not capable of doing: namely, remote professional graphics or other applications that need things like GPU partitioning. These parts look to be aimed at one application only, artificial intelligence, and are meant to process huge amounts of data. For this purpose, their passive coolers will do the job and the lack of ProViz or VDI-oriented drives ensure that AMD will leave these lucrative markets for itself.
As GPU families enter the later part of their lifecycles, we often see chip manufacturers start to offload stockpiles of salvaged chips that, for one reason or another, didn't make the grade for the tier of cards they normally are used in. These recovered chips are fairly unremarkable overall, but they are unsold silicon that still works and has economic value, leading to them being used in lower-tier cards so that they can be sold. And, judging by the appearance of a new video card design from MSI, it looks like NVIDIA's Ada Lovelace generation of chips has reached that stage, as the Taiwanese video card maker has put out a new GeForce RTX 4070 Ti Super card based on a salvaged AD102 GPU.
Typically based on NVIDIA's AD103 GPU, NVIDIA's GeForce RTX 4070 Ti Super series sits a step below the company's flagship RTX 4080/4090 cards, both of which are based on the bigger and badder AD102 chip. But with some number of AD102 chips inevitably failing to live up to RTX 4080 specifications, rather than being thrown out, these chips can instead be used to make RTX 4070 cards. Which is exactly what MSI has done with their new GeForce RTX 4070 Ti Super Ventus 3X Black OC graphics card.
The card itself is relatively unremarkable – using a binned AD102 chip doesn't come with any advantages, and it should perform just like regular AD103 cards – and for that reason, video card vendors rarely publicly note when they're doing a run of cards with a binned-down version of a bigger chip. However, these larger chips have a tell-tale PCB footprint that usually makes it obvious what's going on. Which, as first noticed by @wxnod, is exactly what's going on with MSI's card.
Ada Lovelace Lineup: MSI GeForce RTX 4070 TiS (AD103), RTX 4070 TiS (AD102), & RTX 4090 (AD102)
The tell, in this case, is the rear board shot provided by MSI. The larger AD102 GPU uses an equally larger mounting bracket, and is paired with a slightly more complex array of filtering capacitors on the back side of the board PCB. Ultimately, since these are visible in MSI's photos of their GeForce RTX 4070 Ti Super Ventus 3X Black OC, it's easy to compare it to other video cards and see that it has exactly the same capacitor layout as MSI's GeForce RTX 4090, thus confirming the use of an AD102 GPU.
Chip curiosities aside, all of NVIDIA GeForce RTX 4070 Ti Super graphics cards – no matter whether they are based on the AD102 or AD103 GPU – come with a GPU with 8,448 active CUDA cores and 16 GB of GDDR6X memory, so it doesn't (typically) matter which chip they carry. Otherwise, compared to a fully-enabled AD102 chip, the RTX 4070 Ti Super specifications are relatively modest, with fewer than half as many CUDA cores, underscoring how the AD102 chip being used in MSI's card is a pretty heavy salvage bin.
As for the rest of the card, MSI GeForce RTX 4070 Ti Super Ventus 3X Black OC is a relatively hefty card overall, with a cooling system to match. Being overclocked, the Ventus also has a slightly higher TDP than normal GeForce RTX 4070 Ti Super cards, weighing in at 295 Watts, or 10 Watts above baseline cards.
Meanwhile, MSI is apparently not the only video card manufacturer using salvaged AD102 chips for GeForce RTX 4070 Ti Super, either. @wxnod has also posted a screenshot obtained on an Inno3D GeForce RTX 4070 Ti Super based on an AD102 GPU.
Thanks to the success of the burgeoning market for AI accelerators, NVIDIA has been on a tear this year. And the only place that’s even more apparent than the company’s rapidly growing revenues is in the company’s stock price and market capitalization. After breaking into the top 5 most valuable companies only earlier this year, NVIDIA has reached the apex of Wall Street, closing out today as the world’s most valuable company.
With a closing price of $135.58 on a day that saw NVIDIA’s stock pop up another 3.5%, NVIDIA has topped both Microsoft and Apple in valuation, reaching a market capitalization of $3.335 trillion. This follows a rapid rise in the company’s stock price, which has increased by 47% in the last month alone – particularly on the back of NVIDIA’s most recent estimates-beating earnings report – as well as a recent 10-for-1 stock split. And looking at the company’s performance over a longer time period, NVIDIA’s stock jumped a staggering 218% over the last year, or a mere 3,474% over the last 5 years.
NVIDIA’s ascension continues a trend over the last several years of tech companies all holding the top spots in the market capitalization rankings. Though this is the first time in quite a while that the traditional tech leaders of Apple and Microsoft have been pushed aside.
| Market Capitalization Rankings | ||
| Market Cap | Stock Price | |
| NVIDIA | $3.335T | $135.58 |
| Microsoft | $3.317T | $446.34 |
| Apple | $3.285T | $214.29 |
| Alphabet | $2.170T | $176.45 |
| Amazon | $1.902T | $182.81 |
Driving the rapid growth of NVIDIA and its market capitalization has been demand for AI accelerators from NVIDIA, particularly the company’s server-grade H100, H200, and GH200 accelerators for AI training. As the demand for these products has spiked, NVIDIA has been scaling up accordingly, repeatedly beating market expectations for how many of the accelerators they can ship – and what price they can charge. And despite all that growth, orders for NVIDIA’s high-end accelerators are still backlogged, underscoring how NVIDIA still isn’t meeting the full demands of hyperscalers and other enterprises.
Consequently, NVIDIA’s stock price and market capitalization have been on a tear on the basis of these future expectations. With a price-to-earnings (P/E) ratio of 76.7 – more than twice that of Microsoft or Apple – NVIDIA is priced more like a start-up than a 30-year-old tech company. But then it goes without saying that most 30-year-old tech companies aren’t tripling their revenue in a single year, placing NVIDIA in a rather unique situation at this time.
Like the stock market itself, market capitalizations are highly volatile. And historically speaking, it’s far from guaranteed that NVIDIA will be able to hold the top spot for long, never mind day-to-day fluctuations. NVIDIA, Apple, and Microsoft’s valuations are all within $50 billion (1.%) of each other, so for the moment at least, it’s still a tight race between all three companies. But no matter what happens from here, NVIDIA gets the exceptionally rare claim of having been the most valuable company in the world at some point.
(Carousel image courtesy MSN Money)
Update 06/13: SK hynix has sent a note to AnandTech clarifying that the company "plans to start mass production of GDDR7 in the fourth quarter of this year when the relevant market opens up." This article has been updated accordingly.
Being a major JEDEC memory standard, GDDR7 is slated to be produced by all three of the Big Three memory manufacturers. But it seems that not all three vendors will be kicking off mass production at the same time.
SK hynix was at this year's Computex trade show, showing off their full lineup of memory technologies – including, of course, GDDR7. SK hynix is the last of the major memory vendor's we've seen promoting their memory, and fittingly, they seem to be the last in terms of their mass production schedule. According to company representatives, the firm will kick off mass production of their GDDR7 chips in the last quarter of 2024.
Comparatively, the company's cross-town rival, Samsung, is already sampling memory with the goal of getting it out the door in 2024. And Micron has been rather gung ho about not only starting mass production this year, but starting it early enough that at least some of their customers will be able to ship finished products this year.
That said, it bears mentioning that with industry-standard memory technologies, mass production at one vendor does not indicate that another is late; it is just indicating that someone was first to validate with a partner and that partner plans to ship its product in 2024. And while mass production remains another 4+ months out, SK hynix does have sample chips for its partners to test right now, and the chips have been demonstrated at Computex.
As far as SK hynix's floor booth at Computex 2024 is concerned, the company had GDDR7 chips on display along with a table essentially summarizing the company's roadmap. For now, SK hynix is planning on both 16Gbit and 24Gbit chips, with data transfer rates of up to 40 GT/s. Though when SK hynix intends to launch their higher-end configurations remains to be seen. Both of the company's rivals are starting out with 16Gbit chips running at 32 GT/sec, so being the first to get a faster/larger chip out would be a feather in SK hynix's cap.
While the bulk of AMD’s Computex presentation was on CPUs and their Instinct lineup of dedicated AI accelerators, the company also has a small product refresh for the professional graphics and workstation AI crowd. AMD is releasing a dual-slot version of their high-end Radeon Pro W7900 card – aptly named the W7900 Dual Slot – with the intent being to improve compute density in workstations by making it possible to install 4 of the cards inside a single chassis.
The release of a dual-slot version of the card comes after the original Radeon Pro W7900 was the first time AMD went with a larger, triple-slot form factor for their flagship workstation card. With the W7000 generation bringing an all-around increase in power consumption, pushing the W7900 to 295 Watts, AMD originally opted to release a larger card for improved acoustics. However this came at the cost of compute density, as most systems could only fit 2 of the thicker cards. As a result, AMD is opting to release a dual-slot version of the hardware as well, to offer a more competitive product for high-density workstation systems – particularly those doing local AI inference.
| AMD Radeon Pro Specification Comparison | ||||||
| AMD Radeon Pro W7900DS | AMD Radeon Pro W7900 | AMD Radeon Pro W7800 | AMD Radeon Pro W6800 | |||
| ALUs | 12288 (96 CUs) |
8960 (70 CUs) |
3840 (60 CUs) |
|||
| ROPs | 192 | 128 | 96 | |||
| Boost Clock | 2.495GHz | 2.495GHz | 2.32HHz | |||
| Peak Throughput (FP32) | 61.3 TFLOPS | 45.2 TFLOPS | 17.8 TFLOPS | |||
| Memory Clock | 18 Gbps GDDR6 | 18 Gbps GDDR6 | 16 Gbps GDDR6 | |||
| Memory Bus Width | 384-bit | 256-bit | 256-bit | |||
| Memiry Bandwidth | 864GB/sec | 576GB/sec | 512GB/sec | |||
| VRAM | 48GB | 32GB | 32GB | |||
| ECC | Yes (DRAM) |
Yes (DRAM) |
Yes (DRAM) |
|||
| Infinity Cache | 96MB | 64MB | 128MB | |||
| Total Board Power | 295W | 260W | 250W | |||
| Manufacturing Process | GCD: TSMC 5nm MCD: TSMC 6nm |
GCD: TSMC 5nm MCD: TSMC 6nm |
TSMC 7nm | |||
| Architecture | RDNA3 | RDNA3 | RDNA2 | |||
| GPU | Navi 31 | Navi 31 | Navi 21 | |||
| Form Factor | Dual Slot Blower | Triple Slot Blower | Dual Slot Blower | Dual Slot Blower | ||
| Launch Date | 06/2024 | Q2'2023 | Q2'2023 | 06/2021 | ||
| Launch Price (MSRP) | $3499 | $3999 | $2499 | $2249 | ||
Other than the narrower cooler, the Radeon Pro W7900DS is for all intents and purposes identical to the original W7900, with the same Navi 31 GPU being driven to the same clockspeeds, and the overall board being run to the same 295 Total Board Power (TBP) limit. This is paired with the same 18Gbps GDDR6 as before, giving the card 48GB of VRAM.
Officially, AMD doesn’t have a noise specification for these cards. But you can expect that the W7900DS will be louder than its triple-slot senior. By all appearances, AMD is just using the cooler from the W7800, which was a dual-slot card from the start, so that cooler is being tasked with handling another 35W of heat dissipation.
As the W7800 was also AMD’s fastest dual-slot card up until now, it’s an apt point of comparison for compute density. With its full-fat Navi 31 GPU, the W7900DS will offer about 36% more compute/pixel throughput than its sibling/predecessor. So it’s a not-insubstantial improvement for the very specific niche AMD has in mind for the card.
And like so many other things being announced at Computex this year, that niche is AI. While AMD offers PCIe versions of their Instinct MI210 accelerators, those cards are geared at servers, with fully-passive coolers to match. So workstation-level compute is largely picked up by AMD’s Radeon Pro workstation cards, which are intended to go into a traditional PC chassis and use active cooling (blowers). In this case, AMD is specifically going after local inference workloads, as that’s what the Radeon hardware and its significant VRAM pool are best suited for.
The Radeon Pro W7900 Dual Slot will drop on June 19th. Notably, AMD is introducing the card at a slightly lower price tag than they launched the original W7900 at last year, with the W7900DS hitting retail shelves at $3499, down from the W7900’s original $3999 price tag.
Alongside the release of the W7900DS, AMD is also promoting the upcoming Radeon release of ROCm 6.1, their software stack for GPU computing. While baseline ROCm 6.1 was introduced back in April, the Windows version of AMD’s software stack is still a trailing (and feature limited) release. So that is slated to finally get bumped up to a ROCm 6.1 release on June 19th, the same day the W7900DS launches.
ROCm 6.1 for Radeons is slated to bring a couple of major changes/improvements to the stack, particularly when it comes to expanding the scope of available features. Notably, AMD will finally be shipping Windows Subsystem for Linux 2 (WSL2) support, albeit at a beta level, allowing Windows users to access the much richer feature set and software ecosystem of ROCm under Linux. This release will also incorporate improved support for multi-GPU configurations, perfect timing for the launch of the Radeon Pro W7900DS.
Finally, ROCm 6.1 sees TensorFlow integrated into the ROCm software stack as a first-class citizen. While this matter involves more complexities than can be summarized in a simple news story, native TensorFlow support under Windows was previously blocked by a lack of a Windows version of AMD’s MIOpen machine learning library. Combined with WSL2 support, developers will have two ways to access TensorFlow on Windows systems going forward.
In a packed presentation kicking off this year’s Computex trade show, AMD CEO Dr. Lisa Su spent plenty of time focusing on the subject of AI. And while the bulk of that focus was on AMD’s impending client products, the company is also currently enjoying the rapid growth of their Instinct lineup of accelerators, with the MI300 continuing to break sales projections and growth records quarter after quarter. It’s no surprise then that AMD is looking to move quickly then in the AI accelerator space, both to capitalize on the market opportunities amidst the current AI mania, as well as to stay competitive with the many chipmakers large and small who are also trying to stake a claim in the space.
To that end, as part of this evening’s announcements, AMD laid out their roadmap for their Instinct product lineup for both the short and long term, with new products and new architectures in development to carry AMD through 2026 and beyond.
On the product side of matters, AMD is announcing a new Instinct accelerator, the HBM3E-equipped MI325X. Based on the same computational silicon as the company’s MI300X accelerator, the MI325X swaps out HBM3 memory for faster and denser HBM3E, allowing AMD to produce accelerators with up to 288GB of memory, and local memory bandwidths hitting 6TB/second.
Meanwhile, AMD also showcased their first new CDNA architecture/Instinct product roadmap in two years, laying out their plans through 2026. Over the next two years AMD will be moving very quickly indeed, launching two new CDNA architectures and associated Instinct products in 2025 and 2026, respectively. The CDNA 4-powered MI350 series will be released in 2025, and that will be followed up by the even more ambitious MI400 series in 2026, which will be based on the CDNA "Next" architecture.
With NVIDIA’s Turing architecture turning six years old this year, the company has been retiring many of the remaining Turing products from its video card lineup. And today that spirit of spring cleaning is coming to the entry-level segment of NVIDIA’s professional visualization lineup, where NVIDIA is introducing a pair of new desktop cards based on their low-end Ampere hardware.
The new RTX A1000 and RTX A400 cards will be replacing the T1000/T600/T400 lineup, which was released three years ago in 2021. The new cards slot into the same entry-level category and finally finish fleshing out the RTX A series of proviz cards, offering NVIDIA’s Ampere-generation professional graphics technologies in the lowest-power, lowest-performance, lowest-cost configuration possible.
Notably, since the entry-level T-series were based on NVIDIA’s feature-limited TU11x silicon, which lacked ray tracing and tensor core support – the basis of NVIDIA’s RTX technologies and associated branding – this marks the first time these technologies will be available in NVIDIA’s entry-level desktop proviz cards. And accordingly, these are being promoted to RTX-branded video cards, ending the odd overlap with NVIDIA’s compute cards, which never carry RTX branding.
It goes without saying that as low-end cards, the ray tracing performance of either part is nothing to write home about, but it gives NVIDIA’s current proviz lineup a consistent set of graphics features from top to bottom.
| NVIDIA Professional Visualization Card Specification Comparison | ||||||
| A1000 | A400 | T1000 | T400 | |||
| CUDA Cores | 2304 | 768 | 896 | 384 | ||
| Tensor Cores | 72 | 24 | N/A | N/A | ||
| Boost Clock | 1460MHz | 1755MHz | 1395MHz | 1425MHz | ||
| Memory Clock | 12Gbps GDDR6 | 12Gbps GDDR6 | 10Gbps GDDR6 | 10Gbps GDDR6 |
||
| Memory Bus Width | 128-bit | 64-bit | 128-bit | 64-bit | ||
| VRAM | 8GB | 4GB | 8GB | 4GB | ||
| Single Precision | 6.74 TFLOPS | 2.7 TFLOPS | 2.5 TFLOPS | 1.09 TFLOPS | ||
| Tensor Performance | 53.8 TFLOPS | 21.7 TFLOPS | N/A | N/A | ||
| TDP | 50W | 50W | 50W | 30W | ||
| Cooling | Active, SS | Active, SS | Active, SS | Active, SS | ||
| Outputs | 4x mDP 1.4a | 4x mDP 1.4a | 3x mDP 1.4a | |||
| GPU | GA107 | TU117 | ||||
| Architecture | Ampere | Turing | ||||
| Manufacturing Process | Samsung 8nm | TSMC 12nm | ||||
| Launch Date | 04/2024 | 05/2024 | 05/2021 | 05/2021 | ||
Both the A1000 and A400 are based on the same board design, with NVIDIA doing away with any pretense of physical feature differentiation this time around (T400 was missing its 4th Mini DisplayPort). This means both cards are based on the GA107 GPU, sporting different core and memory configurations.
RTX A1000 is a not-quite-complete configuration of GA107, with 2304 CUDA cores and 72 tensor cores. This is paired with 8GB of GDDR6, which runs at 12Gbps, for a total of 192GB/second of memory bandwidth. The TDP of the card is 50 Watts, matching its predecessor.
Meanwhile RTX A400 is far more cut down, offering about a third of the active hardware on the GPU itself, and half the memory bandwidth. On paper this gives it around 40% of T1000’s performance, and half the memory bandwidth – or 96GB/second. Notably, despite the hardware cut-down, the official TDP is still 50 Watts, versus the 30 Watts of its predecessor. So at this point NVIDIA will soon cease offering a desktop proviz card lower than 50 Watts.
As noted before, both cards otherwise feature the same physical design, with a half-height half-length (HHHL) board with active cooling. As you’d expect from such low-TDP cards, these are single-slot cooler designs. Both cards feature a quartet of Mini DisplayPorts, with the same DP 1.4a functionality that we’ve seen across all of NVIDIA’s products for the last several years.
Finally, video-focused users will want to make note that the A1000/A400 have slightly different video capabilities. While A1000 gets access to both of GA107’s NVDEC video decode blocks, A400 only gets access to a single block – one more cutback to differentiate the two cards. Otherwise, both video cards get access to the GPU’s sole NVENC block.
According to NVIDIA, the RTX A1000 will be available starting today through its distribution partners. Meanwhile the RTX A400 will hit distribution channels in May, and with OEMs expected to begin offering the cards as part of their pre-built systems this summer.
Intel this morning is kicking off the second day of their Vision 2024 conference, the company’s annual closed-door business and customer-focused get-together. While Vision is not typically a hotbed for new silicon announcements from Intel – that’s more of an Innovation thing in the fall – attendees of this year’s show are not coming away empty handed. With a heavy focus on AI going on across the industry, Intel is using this year’s event to formally introduce the Gaudi 3 accelerator, the next-generation of Gaudi high-performance AI accelerators from Intel’s Habana Labs subsidiary.
The latest iteration of Gaudi will be launching in the third quarter of 2024, and Intel is already shipping samples to customers now. The hardware itself is something of a mixed bag in some respects (more on that in a second), but with 1835 TFLOPS of FP8 compute throughput, Intel believes it’s going to be more than enough to carve off a piece of the expansive (and expensive) AI market for themselves. Based on their internal benchmarks, the company expects to be able beat NVIDIA’s flagship Hx00 Hopper architecture accelerators in at least some critical large language models, which will open the door to Intel grabbing a larger piece of the AI accelerator market at a critical time in the industry, and a moment when there simply isn’t enough NVIDIA hardware to go around.
Introspect this week introduced its M5512 GDDR7 memory test system, which is designed for testing GDDR7 memory controllers, physical interface, and GDDR7 SGRAM chips. The tool will enable memory and processor manufacturers to verify that their products perform as specified by the standard.
One of the crucial phases of a processor design bring up is testing its standard interfaces, such as PCIe, DisplayPort, or GDDR is to ensure that they behave as specified both logically and electrically and achieve designated performance. Introspect's M5512 GDDR7 memory test system is designed to do just that: test new GDDR7 memory devices, troubleshoot protocol issues, assess signal integrity, and conduct comprehensive memory read/write stress tests.
The product will be quite useful for designers of GPUs/SoCs, graphics cards, PCs, network equipment and memory chips, which will speed up development of actual products that rely on GDDR7 memory. For now, GPU and SoC designers as well as memory makers use highly-custom setups consisting of many tools to characterize signal integrity as well as conduct detailed memory read/write functional stress testing, which are important things at this phase of development. But usage of a single tool greatly speeds up all the processes and gives a more comprehensive picture to specialists.
The M5512 GDDR7 Memory Test System is a desktop testing and measurement device that is equippped with 72 pins capable of functioning at up to 40 Gbps in PAM3 mode, as well as offering a virtual GDDR7 memory controller. The device features bidirectional circuitry for executing read and write operations, and every pin is equipped with an extensive range of analog characterization features, such as skew injection with femto-second resolution, voltage control with millivolt resolution, programmable jitter injection, and various eye margining features critical for AC characterization and conformance testing. Furthermore, the system integrates device power supplies with precise power sequencing and ramping controls, providing a comprehensive solution for both AC characterization and memory functional stress testing on any GDDR7 device.
Introspects M5512 has been designed in close collaboration with JEDEC members working on the GDDR7 specification, so it promises to meet all of their requirements for compliance testing. Notably, however, the device does not eliminate need for interoperability tests and still requires companies to develop their own test algorithms, but it's still a significant tool for bootstrapping device development and getting it to the point where chips can begin interop testing.
“In its quest to support the industry on GDDR7 deployment, Introspect Technology has worked tirelessly in the last few years with JEDEC members to develop the M5512 GDDR7 Memory Test System,” said Dr. Mohamed Hafed, CEO at Introspect Technology.
AMD's FidelityFX Super Resolution 3 technology package introduced a plethora of enhancements to the FSR technology on Radeon RX 6000 and 7000-series graphics cards last September. But perfection has no limits, so this week, the company is rolling out its FSR 3.1 technology, which improves upscaling quality, decouples frame generation from AMD's upscaling, and makes it easier for developers to work with FSR.
Arguably, AMD's FSR 3.1's primary enhancement is its improved temporal upscaling image quality: compared to FSR 2.2, the image flickers less at rest and no longer ghosts when in movement. This is a significant improvement, as flickering and ghosting artifacts are particularly annoying. Meanwhile, FSR 3.1 has to be implemented by the game developer itself, and the first title to support this new technology sometime later this year is Ratchet & Clank: Rift Apart.
| Temporal Stability | |
|
|
|
| AMD FSR 2.2 | AMD FSR 3.1 |
| Ghosting Reduction | |
|
|
|
| AMD FSR 2.2 | AMD FSR 3.1 |
Another significant development brought by FSR 3.1 is its decoupling from the Frame Generation feature introduced by FSR 3. This capability relies on a form of AMD's Fluid Motion Frames (AFMF) optical flow interpolation. It uses temporal game data like motion vectors to add an additional frame between existing ones. This ability can lead to a performance boost of up to two times in compatible games, but it was initially tied to FSR 3 upscaling, which is a limitation. Starting from FSR 3.1, it will work with other upscaling methods, though AMD refrains from saying which methods and on which hardware for now. Also, the company does not disclose when it is expected to be implemented by game developers.
In addition, AMD is bringing support for FSR3 to Vulkan and Xbox Game Development Kit, enabling game developers on these platforms to use it. It also adds FSR 3.1 to the FidelityFX API, which simplifies debugging and enables forward compatibility with updated versions of FSR.
Upon its release in September 2023, AMD FSR 3 was initially supported by two titles, Forspoken and Immortals of Aveum, with ten more games poised to join them back then. Fast forward to six months later, the lineup has expanded to an impressive roster of 40 games either currently supporting or set to incorporate FSR 3 shortly. As of March 2024, FSR is supported by games like Avatar: Frontiers of Pandora, Starfield, The Last of Us Part I. Shortly, Cyberpunk 2077, Dying Light 2 Stay Human, Frostpunk 2, and Ratchet & Clank: Rift Apart will support FSR shortly.
Source: AMD
Already solidly in the driver’s seat of the generative AI accelerator market at this time, NVIDIA has long made it clear that the company isn’t about to slow down and check out the view. Instead, NVIDIA intends to continue iterating along its multi-generational product roadmap for GPUs and accelerators, to leverage its early advantage and stay ahead of its ever-growing coterie of competitors in the accelerator market. So while NVIDIA’s ridiculously popular H100/H200/GH200 series of accelerators are already the hottest ticket in Silicon Valley, it’s already time to talk about the next generation accelerator architecture to feed NVIDIA’s AI ambitions: Blackwell.
Asus this week has become the latest PC video card manufacturer to announce a sub-75W video card based on NVIDIA's recently-released low-power GeForce RTX 3050 6GB design. And going one step further for small form factor PC owners, Asus has used NVIDIA's low-power GPU configuration to produce a half-height video card that can fit into low-profile systems.
As Asus puts it, the GeForce RTX 3050 LP BRK 6GB GDDR6 is a 'big productivity in a small package' and for a low-profile dual-slot graphics board, it indeed is. The unit has three display outputs, including a DVI-D, HDMI 2.1, and DisplayPort 1.4a with HDCP 2.3 support, which makes the graphics card s viable option both for a a dual-display desktop and a home theater PC (Nvidia's GA107 graphics processor supports all popular codecs except AV1). Furthermore, a DVI-D output enables the card to drive outdated displays, which even over half a decade after DVI-D was retired, still hang around as spare parts. Meanwhile, because the card only consumes around 70W, it does not require any auxiliary PCIe power connectors, which are at times not available in cheap systems from big PC makers.
Underlying this card is the aforementioned GeForce RTX 3050 6 GB, which uses the GA107 GPU with 2304 CUDA cores, and it comes with 6GB of GDDR6 memory connected to a narrower 96-bit memory bus (down from 128-bits for the full 8GB version. With a lower boost clock of 1470 MHz (1500 MHz in OC mode), the RTX 3050 6GB has reduced computing performance, delivering 6.77 FP32 TFLOPS versus 9.1 FP32 TFLOPS of the full-fledged RTX 3050.
As a result, the low-profile GeForce RTX 3050 6 GB is very much an entry-level card, though the low power requirements for such a card are also what make it special. This should be plenty for low-end gaming – beating out integrated GPUs – though suffice it to say, it's not going to compete with high-end, power-hungry cards either.
With its diminutive size, the Asus GeForce RTX 3050 LP BRK 6 GB GDDR6 looks to be a nice candidate for upgrading cheap systems from OEMs as well as fixing outdated PCs. What remains to be seen is how price competitive it is going to be. The graphics board already has one low-profile rival from MSI — which costs $185 — so Asus is not the only vendor competing here.
JEDEC on Tuesday published the official specifications for GDDR7 DRAM, the latest iteration of the long-standing memory standard for graphics cards and other GPU-powered devices. The newest generation of GDDR brings a combination of memory capacity and memory bandwidth gains, with the later being driven primarily by the switch to PAM3 signaling on the memory bus. The latest graphics RAM standard also boosts the number of channels per DRAM chip, adds new interface training patterns, and brings in on-die ECC to maintain the effective reliability of the memory.
“JESD239 GDDR7 marks a substantial advancement in high-speed memory design,” said Mian Quddus, JEDEC Board of Directors Chairman. “With the shift to PAM3 signaling, the memory industry has a new path to extend the performance of GDDR devices and drive the ongoing evolution of graphics and various high-performance applications.”
GDDR7 has been in development for a few years now, with JEDEC members making the first disclosures around the memory technology about a year ago, when Cadence revealed the use of PAM3 encoding as part of their validation tools. Since then we've heard from multiple memory manufacturers that we should expect the final version of the memory to launch in 2024, with JEDEC's announcement essentially coming right on schedule.
As previously revealed, the biggest technical change with GDDR7 comes with the switch from two-bit non-return-to-zero (NRZ) encoding on the memory bus to three-bit pulse amplitude modulating (PAM3) encoding. This change allows GDDR7 to transmit 3 bits over two cycles, 50% more data than GDDR6 operating at an identical clockspeed. As a result, GDDR7 can support higher overall data transfer rates, the critical component to making each generation of GDDR successively faster than its predecessor.
| GDDR Generations | |||||
| GDDR7 | GDDR6X (Non-JEDEC) |
GDDR6 | |||
| B/W Per Pin | 32 Gbps (Gen 1) 48 Gbps (Spec Max) |
24 Gbps (Shipping) | 24 Gbps (Sampling) | ||
| Chip Density | 2 GB (16 Gb) | 2 GB (16 Gb) | 2 GB (16 Gb) | ||
| Total B/W (256-bit bus) | 1024 GB/sec | 768 GB/sec | 768 GB/sec | ||
| DRAM Voltage | 1.2 V | 1.35 V | 1.35 V | ||
| Data Rate | QDR | QDR | QDR | ||
| Signaling | PAM-3 | PAM-4 | NRZ (Binary) | ||
| Maximum Density | 64 Gb | 32 Gb | 32 Gb | ||
| Packaging | 266 FBGA | 180 FBGA | 180 FBGA | ||
The first generation of GDDR7 is expected to run at data rates around 32 Gbps per pin, and memory manufacturers have previously talked about rates up to 36 Gbps/pin as being easily attainable. However the GDDR7 standard itself leaves room for even higher data rates – up to 48 Gbps/pin – with JEDEC going so far as touting GDDR7 memory chips "reaching up to 192 GB/s [32b @ 48Gbps] per device" in their press release. Notably, this is a significantly higher increase in bandwidth than what PAM3 signaling brings on its own, which means there are multiple levels of enhancements within GDDR7's design.
Digging deeper into the specification, JEDEC has also once again subdivided a single 32-bit GDDR memory chip into a larger number of channels. Whereas GDDR6 offered two 16-bit channels, GDDR7 expands this to four 8-bit channels. The distinction is somewhat arbitrary from an end-user's point of view – it's still a 32-bit chip operating at 32Gbps/pin regardless – but it has a great deal of impact on how the chip works internally. Especially as JEDEC has kept the 256-bit per channel prefetch of GDDR5 and GDDR6, making GDDR7 a 32n prefetch design.
GDDR Channel Architecture. Original GDDR6-era Diagram Courtesy Micron
The net impact of all of this is that, by halving the channel width but keeping the prefetch size the same, JEDEC has effectively doubled the amount of data that is prefetched per cycle of the DRAM cells. This is a pretty standard trick to extend the bandwidth of DRAM memory, and is essentially the same thing JEDEC did with GDDR6 in 2018. But it serves as a reminder that DRAM cells are still very slow (on the order of hundreds of MHz) and aren't getting any faster. So the only way to feed faster memory buses is by fetching ever-larger amounts of data in a single go.
The change in the number of channels per memory chip also has a minor impact on how multi-channel "clamshell" mode works for higher capacity memory configurations. Whereas GDDR6 accessed a single memory channel from each chip in a clamshell configuration, GDDR7 will access two channels – what JEDEC is calling two-channel mode. Specifically, this mode reads channels A and C from each chip. It is effectively identical to how clamshell mode behaved with GDDR6, and it means that while clamshell configurations remain supported in this latest generation of memory, there aren't any other tricks being employed to improve memory capacity beyond ever-increasing memory chip densities.
On that note, the GDDR7 standard officially adds support for 64Gbit DRAM devices, twice the 32Gbit max capacity of GDDR6/GDDR6X. Non-power-of-two capacities continue to be supported as well, allowing for 24Gbit and 48Gbit chips. Support for larger memory chips further pushes the maximum memory capacity of a theoretical high-end video card with a 384-bit memory bus to as high as 192GB of memory – a development that would no doubt be welcomed by datacenter operators in the era of large language AI models. With that said, however, we're still regularly seeing 16Gbit memory chips used on today's memory cards, even though GDDR6 supports 32Gbit chips. Coupled with the fact that Samsung and Micron have already disclosed that their first generation of GDDR7 chips will also top out at 16Gbit/24Gbit respectively, it's safe to say that 64Gbit chips are pretty far off in the future right now (so don't sell off your 48GB cards quite yet).
For their latest generation of memory technology, JEDEC is also including several new-to-GDDR memory reliability features. Most notably, on-die ECC capabilities, similar to what we saw with the introduction of DDR5. And while we haven't been able to get an official comment from JEDEC on why they've opted to include ECC support now, its inclusion is not surprising given the reliability requirements for DDR5. In short, as memory chip densities have increased, it has become increasingly hard to yield a "perfect" die with no flaws; so adding on-chip ECC allows memory manufacturers to keep their chips operating reliably in the face of unavoidable errors.
This figure is reproduced, with permission, from JEDEC document JESD239, figure 124
Internally, the GDDR7 spec requires a minimum of 16 bits of parity data per 256 bits of user data (6.25%), with JEDEC giving an example implementation of a 9-bit single error correcting code (SEC) plus a 7-bit cyclic redundancy check (CRC). Overall, GDDR7 on-die ECC should be able to correct 100% of 1-bit errors, and detect 100% of 2-bit errors – falling to 99.3% in the rare case of 3-bit errors. Information about memory errors is also made available to the memory controller, via what JEDEC terms their on-die ECC transparency protocol. And while technically separate from ECC itself, GDDR7 also throws in another memory reliability feature with command address parity with command blocking (CAPARBLK), which is intended to improve the integrity of the command address bus.
Otherwise, while the inclusion of on-die ECC isn't likely to have any more of an impact on consumer video cards than its inclusion had for DDR5 memory and consumer platforms there, it remains to be seen what this will mean for workstation and server video cards. The vendors there have used soft ECC on top of unprotected memory for several generations now; presumably this will remain the case for GDDR7 cards as well, but the regular use of soft ECC makes things a lot more flexible than in the CPU space.
This figure is reproduced, with permission, from JEDEC document JESD239, figure 152
Finally, GDDR7 is also introducing a suite of other reliability-related features, primarily related to helping PAM3 operation. This includes core independent LFSR (linear-feedback shift register) training patterns with eye masking and error counters. LFSR training patterns are used to test and adjust the interface (to ensure efficiency), eye masking evaluates signal quality, and error counters track the number of errors during training.
Technical matters aside, this week's announcement includes statements of support from all of the usual players on both sides of the isle, including AMD and NVIDA, and the Micron/Samsung/SKhynix trifecta. It goes without saying that all parties are keen to getting to use or sell GDDR7 respectively, given the memory capacity and bandwidth improvements it will bring – and especially in this era where anything aimed at the AI market is selling like hotcakes.
No specific products are being announced at this time, but with Samsung and Micron having previously announced their intentions to ship GDDR7 memory this year, we should see new memory (and new GPUs to pair it with) later this year.
JEDEC standards and publications are copyrighted by the JEDEC Solid State Technology Association. All rights reserved.
NVIDIA today is quietly launching a new entry-level graphics card for the retail market, the GeForce RTX 3050 6GB. Based on a cut-down version of their budget Ampere-architecture GA107 GPU, the new card brings what was previously an OEM-only product to the retail market. Besides adding another part to NVIDIA's deep product stack, the launch of the RTX 3050 6GB also comes with another perk: lower power consumption thanks to this part targeting system installs where an external PCIe power connector would not be needed. NVIDIA's partners, in turn, have not wasted any time in taking advantage of this, and today Palit is releasing its first fanless KalmX board in years: the GeForce RTX 3050 KalmX 6GB.
The GeForce RTX 3050 6GB is based on the GA107 graphics processor with 2304 CUDA cores, which is paired with 6GB of GDDR6 attached to a petite 96-bit memory bus (versus 128-bit for the full RTX 3050 8GB). Coupled with a boost clock rating of just 1470 MHz, the RTX 3050 6GB delivers tangibly lower compute performance than the fully-fledged RTX 3050 — 6.77 FP32 TFLOPS vs 9.1 FP32 TFLOPS — but these compromises offer an indisputable advantage: a 70W power target.
Palit is the first company that takes advantage of this reduced power consumption of the GeForce RTX 3050 6 GB, as the company has launched a passively cooled graphics card based on this part, the first in four years. The Palit GeForce RTX 3050 KalmX 6GB (NE63050018JE-1170H) uses a custom printed circuit board (PCB) that not only offers modern DisplayPort 1.4a and HDMI 2.1 outputs, but, as we still see in some entry-level cards, a dual-link DVI-D connector (a first for an Ampere-based graphics card).
The dual-slot passive cooling system with two heat pipes is certainly the main selling point of Palit's GeForce RTX 3050 KalmX 6GB. The product is pretty large though — it measures 166.3×137×38.3 mm — and will not fit into tiny desktops. Still, given the fact that fanless systems are usually not the most compact ones, this may not be a significant limitation of the new KalmX device.
Another advantage of Palit's GeForce RTX 3050 KalmX 6GB in particular and NVIDIA's GeForce RTX 3050 6GB in general is that it can be powered entirely via a PCIe slot, which eliminates the need for an auxiliary PCIe power connectors (which are sometimes not present in cheap systems from big OEMs).
Wccftech reports that NVIDIA's GeForce RTX 3050 6GB graphics cards will carry a recommended price tag of $169 and indeed these cards are available for $170 - $180. This looks to be a quite competitive price point as the product offers higher compute performance than that of AMD's Radeon RX 6400 ($125) and Radeon RX 6500 XT ($140). Meanwhile, it remains to be seen how much will Palit charge for its uniquely positioned GeForce RTX 3050 KalmX 6GB.
In response to the launch of NVIDIA's new GeForce RTX 40 Super video cards, AMD has announced that they are instituting new promotional pricing on a handful of their high-end video cards in order to keep pace with NVIDIA's new pricing.
Kicking off what AMD is terming a special "promotional pricing" program for the quarter, AMD has been working with its retail partners to bring down the price of the Radeon RX 7900 XT to $749 (or lower), roughly $50 below its street price at the start of the month. Better still, AMD's board partners have already reduced prices further than AMD's official program/projections, and we're seeing RX 7900 XTs drop to as low as $710 in the U.S., making for a $90 drop from where prices stood a few weeks ago.
Meanwhile, AMD is also technically bringing down prices on the China and OEM-only Radeon RX 7900 GRE as well. Though as this isn't available for stand-alone purchase on North American shelves, it's mostly only of relevance for OEM pre-builds (and the mark-ups they charge).
Ultimately, the fact that this is "promotional pricing" should be underscored. The new pricing on the RX 7900 XT is not, at least for the moment, a permanent price cut. Meaning that AMD is leaving themselves some formal room to raise prices later on, if they choose to. Though in practice, it would be surprising to see card prices rebound – at least so long as we don't get a new crypto boom or the like.
Finally, to sweeten the pot, AMD is also extending their latest game bundle offer for another few weeks. The company is offering a copy of Avatar: Frontiers of Pandora with all Radeon RX 7000 video cards (and select Ryzen 7000 CPUs) through January 30, 2024.
NVIDIA has quietly launched its RTX 5880 Ada Generation graphics card that is designed for professional graphics applications. The product is designed to be compliant with the latest U.S. export regulations for China and can be shipped to the People's Republic without restrictions. Meanwhile, it is set to be available globally and sit between the expensive flagship RTX 6000 Ada Generation and considerably less capable RTX 5000 Ada Generation. But there is a major catch about this product.
NVIDIA's RTX 5880 Ada Generation is based on the company's flagship AD102 graphics processor with 14,080 CUDA cores, 110 RT cores, 440 tensor cores, and a 384-bit memory interface to connect 48 GB of GDDR6 memory with ECC. The board comes with four DisplayPort 1.4a connectors and can support either four 4K monitors at 120 Hz, four 5K displays at 60 Hz, or two 8K monitors at 60 Hz. As for power consumption and heat dissipation, it is rated at 285W (with power delivered using one 12VHPWR connector) and comes with a standard dual-slot cooling system with a blower.
As the model number suggests, NVIDIA's RTX 5880 Ada Generation 48 GB should be close to the range-topping RTX 6000 Ada 48 GB. Unfortunately, this is not the case and this is a major catch about the RTX 5880 Ada. From performance point of view, NVIDIA's RTX 5880 Ada offers 69.3 FP32 TFLOPS and 554 FP8 TFLOPS, which is closer to the RTX 5800 Ada 32 GB (65.3 FP32 TFLOPS and 522.1 FP8 TFLOPS) rather than to the RTX 6000 Ada 48 GB (91.1 FP32 TFLOPS and 729 FP8 TFLOPS).
NVIDIA's RTX 6000 Ada has a rather whopping FP8 Total Processing Performance (TPP) score of 5,828 (listed processing power multiplied by the length of operation), which is way higher than 4800 points that the U.S. Department of Commerce wants Chinese entities to get. So, the RTX 5880 Ada has a TPP score of 4432, which comfortably fits within the export requirements of the U.S. government.
But this also means that the RTX 5880 Ada is significantly slower than the RTX 6000 Ada, despite the model number. Truth to be told, the RTX 5880 Ada looks more like an RTX 5000 Ada on steroids rather than a slightly downgraded RTX 6000 Ada.
| NVIDIA's Ada Generation High-End Professional Graphics Cards | |||
| RTX 6000 Ada | RTX 5880 Ada | RTX 5000 Ada | |
| GPU | AD102 | ||
| Memory | 48 GB GDDR6 with ECC | 32 GB GDDR6 ECC | |
| CUDA Cores | 18,176 | 14,080 | 12,800 |
| Tensor Cores | 568 | 440 | 400 |
| RT Cores | 142 | 110 | 100 |
| FP32 Performance | 91.1 TFLOPS | 69.3 TFLOPS | 65.3 TFLOPS |
| RT Performance | 210.6 TFLOPS | 160.2 TFLOPS | 151 TFLOPS |
| Tensor Performance | 729 TFLOPS | 554 TFLOPS | 522.1 TFLOPS |
| TPP | 5828 | 4432 | 4176 |
| Encode/Decode | 3 x encode, 3 x decode (+AV1 encode and decode) | ||
| Display Connectors | 4 x DisplayPort 1.4a | ||
| Power | 300W | 285W | 250W |
| MSRP | $6999 | ? | $4000 |
What remains to be seen is the price of the RTX 5880 Ada. NVIDIA's RTX 5000 Ada has a list price of $4,000, whereas the RTX 6000 Ada has an MSRP of $6,999. It is logical to expect the RTX 5880 Ada to cost closer to the RTX 5000, but not just slightly higher, because after all it has a rather capable 48 GB memory subsystem.
Sources: NVIDIA, RTX 5880 Ada datasheet
Kicking things off for the GPU space at this year’s CES, AMD is at the show to announce that they’re bringing an additional Radeon RX 7000 series card to their lineup: the Radeon RX 7600XT. Intended as a premium version of their existing 1080p-focused RX 7600, the RX 7600 XT bumps things up with roughly 10% higher clockspeeds, as well as a doubling the total amount of VRAM to 16GB of GDDR6. The upgraded Radeon card will hit retail shelves on January 24th for a similarly premium price of $329.
The start of 2024 finds AMD doing some minor shuffling of their product stack both to cover the latest trends in gaming and AI, as well as to cover perceived gaps in the company’s product lineup. While AMD is just in the middle of their product cycle with their actual GPUs – the mid-tier Navi 32 GPU only launched back in August – the company currently has a decently large gap between the $270 Radeon RX 7600 and $450 RX 7700 XT, which they are opting to fill now with the RX 7600 XT. At the same time, video cards with 8GB of VRAM are becoming increasingly unappealing to buyers, especially the VRAM-hungry AI crowd, so this gives AMD the chance to offer a 16GB version of their basic gaming video card.
This morning is an important one for AMD – perhaps the most important of the year. After almost a year and a half of build-up, and even longer for actual development, AMD is launching their next generation GPU/APU/AI accelerator family, the Instinct MI300 series. Based on AMD's new CDNA 3 architecture, and combining it with AMD's proven Zen 4 cores, AMD will be making a full-court press for the high-end GPU and accelerator market with their new product, aiming to lead in both big-metal HPC as well as the burgeoning market for generative AI training and inference.
Taking the stage for AMD's launch event will be AMD CEO Dr. LIsa Su, as well as a numerous AMD executives and ecosystem partners, to detail, at last, AMD's latest generation GPU architecture, and the many forms it will come in. With both the MI300X accelerator and MI300A APU, AMD is aiming to cover most of the accelerator market, whether clients just need a powerful GPU or a tightly-coupled GPU/CPU pairing.
The stakes for today's announcement are significant. The market for generative AI is all but hardware constrained at the moment, much to the benefit of (and profits for) AMD's rival NVIDIA. So AMD is hoping to capitalize on this moment to cut off a piece – perhaps a very big piece – of the market for generative AI accelerators. AMD has made breaking into the server space their highest priority over the last half-decade, and now, they believe, is their time to take a big piece of the server GPU market.
Join us here at 10am Pacific/1pm Eastern/18:00 UTC for our live blog coverage of AMD's event.
Asus this week formally introduced its ASUS Dual GeForce RTX 4060 Ti SSD OC Edition 8 GB graphics card, an unusual video card that also offers an M.2 SSD slot. First showcased earlier this year, the card is aimed at small form factor systems, giving system builders access to a full-bandwidth M.2 slot for additional storage without having to occupy a second PCIe slot.
Just like other GeForce RTX 4060 Ti graphics cards, the Dual GeForce RTX 4060 Ti SSD is based on NVIDIA's AD106 GPU with 4352 CUDA cores (that can operate at up to 2595 MHz). As AD106 only has 8 PCIe lanes out of the gate, regular RTX 4060 Ti cards do not use the other 8 lanes supplied by a PCIe x16 slot. However for their special SSD card, Asus is taking a different tack, using PCIe bifurcation to directly wire up 4 of the remaining 8 lanes to an M.2 slot (no bridges or switch chips required). The end result being that the GPU gets its usual PCIe 4.0 x8 interface while also giving the M.2 slot a PCIe 5.0 x4 interface.
An M.2 2280 slot is located on the backside of the card and can be easily accessed by the user. The cooling system of the graphics card can contact the SSD using a thermal pad, so the cooler of the Dual GeForce RTX 4060 Ti SSD can cool down the drive to ensure its consistent performance even under high loads.
Compared to a prototype board that Asus demonstrated back in June, Asus has simplified the final product. Specifically, whereas the prototype featured two M.2 slots – one of which required removal of the cooling system – the commercial product only has one M.2 slot on the back that is easy to access.
Internally, the use of PCIe bifurcation does present some additional complications. The host needs to support bifurcation of a single PCIe x16 link down to x8 + x8, which virtually all modern AMD and Intel platforms do (and indeed, this is how most of these systems drive multiple x16 slots). But there are the rare exceptions, particularly with entry-level platforms. And even then, bifurcation at the device level isn't something commonly used in consumer hardware, which is to say it's not always well-tested. To that end, Asus offers a list of its own motherboards that are guaranteed to support its Dual GeForce RTX 4060 Ti SSD graphics card.
Otherwise, the Dual GeForce RTX 4060 Ti SSD OC Edition 8GB is a fairly typical RTX 4060 Ti card. Additional power is supplied via an 8-pin PCIe power connector, and Asus offers the standard 3x DisplayPort 1.4a + 1x HDMI 2.1b video outputs. Meanwhile, cooling for the 2.5-slot wide card is provided by a pair of axial fans.
At this point, Asus has not announced a release date or a suggest price for the card. So it remains to be seen when it will become available.
Although AMD doesn’t have any major HPC announcements for this year’s SC23 show, the company isn’t starting this week empty-handed. For the workstation market the company is introducing its latest (and likely final) Radeon Pro W7000 series card, the Radeon Pro W7700.
Based on AMD’s latest-generation RDNA 3 architecture, the W7700 slots in between the Navi 31-based Radeon Pro W7800 and the Navi 33-based Radeon Pro W7600. As the final member of the lineup, the Navi 32-based card is designed to split the difference between AMD’s high-end and low-end Radeon Pro cards, offering a more modest level of performance for a far more modest price tag. Priced at $999, AMD is looking to hit a sweet spot in terms of hitting the high-end workstation market, edging out more expensive NVIDIA cards with a combination of higher performance and lower pricing.
| AMD Radeon RX 7000 Series Specification Comparison | ||||||
| AMD Radeon Pro W7800 | AMD Radeon Pro W7700 | AMD Radeon Pro W7600 | AMD Radeon Pro W5700 | |||
| ALUs | 8960 (70 CUs) |
6144 (48 CUs) |
4096 (32 CUs) |
2304 (36 CUs) |
||
| ROPs | 128 | 96 | 64 | 64 | ||
| Boost Clock | 2.5GHz | ~2.3GHz | 2.43GHz | 1.9GHz | ||
| Peak Throughput (FP32) | 45 TFLOPS | 28.3 TFLOPS | 19.9 TFLOPS | 8.89 TFLOPS | ||
| Memory Clock | 18 Gbps GDDR6 | 18 Gbps GDDR6 | 18 Gbps GDDR6 | 14 Gbps GDDR6 | ||
| Memory Bus Width | 256-bit | 256-bit | 128-bit | 256-bit | ||
| VRAM | 32GB | 16GB | 8GB | 8GB | ||
| ECC | Yes (DRAM) |
Yes (DRAM) |
No | No | ||
| Infinity Cache | 64MB | 64MB | 32MB | N/A | ||
| Total Board Power | 260W | 190W | 130W | 205W | ||
| Manufacturing Process | GCD: TSMC 5nm MCD: TSMC 6nm |
GCD: TSMC 5nm MCD: TSMC 6nm |
TSMC 6nm | TSMC 7nm | ||
| Architecture | RDNA3 | RDNA3 | RDNA3 | RDNA (1) | ||
| GPU | Navi 31 | Navi 32 | Navi 33 | Navi 10 | ||
| Launch Date | Q2'2023 | 11/13/2023 | Q3'2023 | 11/2019 | ||
| Launch Price (MSRP) | $2499 | $999 | $599 | $799 | ||
Under the hood, the Radeon Pro W7700 looks a lot like a Radeon RX 7800XT that’s been dialed down to make a lighter TDP. We’re looking at a Navi 32 GCD with just 48 CUs enabled, and a more modest peak clockspeed of around 2.3GHz. The total rated throughput of the card is 28.3 TFLOPS FP32, trailing the more powerful (and more power-hungry) consumer cards a bit.
The Radeon Pro W7700 is paired with Navi 32’s full suite of Memory Cache Dies (MCD) however, giving it a better-than-usual ratio of FLOPS to memory bandwidth. With 16GB of GDDR6 attached to a 256-bit memory bus, the card offers 576GB/second of memory bandwidth – and ECC-protected memory, at that, a feature carried over from the high-end Radeon Pro models.
As noted earlier, AMD is aiming a bit more conservatively in terms of power consumption for this card, looking to offer something between the toasty 250W+ flagship cards, and the more petite 130W W7600. The end result is that AMD has set the TBP for the W7700 at 190W, allowing for the construction of an easily-cooled dual slot card.
In terms of features, as the fifth RDNA 3 video card in AMD’s Radeon Pro lineup, the W700 is on par with its peers. Like the flagship parts, we’re looking at a fairly performant version of AMD’s RDNA 3 architecture, with the company looking to leverage various optimizations and new features throughout the card to offer better performance, and for specific tasks, new features. The core architecture includes a significant update to the CUs to improve the total number of compute FLOPS attainable, as well as optimizations to improve the kind of dense matrix math used in AI inference calculations. As well, the ray tracing units have been updated to provide better performance and a larger overall ray rate.
RDNA 3’s marquee video feature of AV1 encoding support also makes it over to the card as well. The W7700 can encode a single AV1 stream at up to 8K60, a very useful feature as AV1 adoption increases – though it doesn’t get the more powerful dual media engine found on the flagship W7800/W7900’s cards.
Meanwhile, AMD’s ace in the hole on the display output side of matters is their support for DisplayPort 2.1 connectivity, something all of NVIDIA’s cards in this generation lack. Still, Pro users intending to take advantage of DP 2.1 will want to pay close attention to the specifications for AMD’s Radeon Pro lineup, as support is not consistent throughout the cards. Whereas the cheaper W7500/W7600 only have UHBR 10 data rate support, the W7700 gets faster UHBR 13.5 support for higher resolution/color depth displays. Still, this is a step down from the W7800/W7900, which can uniquely drive a max-bandwidth UHBR 20 display.
As for the physical W7700 card itself, it is a fairly typical-for-workstations dual slot blower design. AMD is using their standard DisplayPort configuration here, equipping the card with a quartet of full-sized DisplayPorts. In terms of length, the W7700 is a bit shorter than its W7800/W7900 counterparts, though AMD hasn’t sent over the exact dimensions for the card. And while not shown in any of AMD's pictures, the card is fed additional power a single 8-pin PCIe power connector.
Wrapping things up, the Radeon Pro W7700 is getting a hard launch, and will be available today in stand-alone card form from the usual PC parts retailers at an MSRP of $999. Meanwhile, OEMs are expected to start shipping the card in their workstations beginning later this month.
With faster and higher capacity HBM3e memory set to come online early in 2024, NVIDIA has been preparing its current-generation server GPU products to use the new memory. Back in August we saw NVIDIA’s plans to release an HBM3e-equipped version of the Grace Hopper GH200 superchip, and now for the SC23 tradeshow, NVIDIA is announcing their plans to bring to market an updated version of the stand-alone H100 accelerator with HBM3e memory, which the company will be calling the H200.
Like its Grace Hopper counterpart, the purpose of the H200 is to serve as a mid-generation upgrade to the Hx00 product line by rolling out a version of the chip with faster and higher capacity memory. Tapping the HBM3e memory that Micron and others are set to roll out n, NVIDIA will be able to offer accelerators with better real-world performance in memory bandwidth-bound workloads, but also parts that can handle even larger workloads. This stands to be especially helpful in the generative AI space – which has been driving virtually all of the demand for H100 accelerators thus far – as the largest of the large language models can max out the 8GB H100 as it is.
Meanwhile, with HBM3e memory not shipping until next year, NVIDIA has been using the gap to announce HBM3e updated parts at their leisure. Following this summer’s GH200 announcement, it was only a matter of time until NVIDIA announced a standalone version of the Hx00 accelerator with HBM3e, and this week NVIDIA is finally making that announcement.
As AMD is now well into their third generation of RDNA architecture GPUs, the sun has been slowly setting on AMD’s remaining Graphics Core Next (GCN) designs, better known by the architecture names of Polaris and Vega. In recent weeks the company dropped support for those GPU architectures in their open source Vulkan Linux driver, AMDVLK, and now we have confirmation that the company is slowly winding down support for these architectures in their Windows drivers as well.
Under AMD’s extended driver support schedule for Polaris and Vega, the drivers for these architectures will no longer be kept at feature parity with the RDNA architectures. And while AMD will continue to support Polaris and Vega for some time to come, that support is being reduced to security updates and “functionality updates as available.”
For AMD users keeping a close eye on their driver releases, they’ll likely recognize that AMD already began this process back in September – though AMD hasn’t officially documented the change until now. As of AMD’s September Adrenaline 23.9 driver series, AMD split up the RDNA and GCN driver packages, and with that they have also split the driver branches between the two architectures. As a result, only RDNA cards are receiving new features and updates as part of AMD’s mainline driver branch (currently 23.20), while the GCN cards have been parked on a maintenance driver branch – 23.19.
At present, AMD has not published anything about this change in driver support to their website. But, responding to a request for a comment on Windows driver support from AnandTech, the company provided the following statement:
The AMD Polaris and Vega graphics architectures are mature, stable and performant and don’t benefit as much from regular software tuning. Going forward, AMD is providing critical updates for Polaris- and Vega-based products via a separate driver package, including important security and functionality updates as available. The committed support is greater than for products AMD categorizes as legacy, and gamers can still enjoy their favorite games on Polaris and Vega-based products.
Notably, AMD is actively asserting that this is not “legacy” status for Vega and Polaris, which is an important distinction because of what “legacy” means within AMD’s ecosystem. For AMD, legacy products are effectively considered end-of-life, and ongoing driver support is retired. Which in the case of previous generation GPU architectures going legacy, AMD did not have any further driver releases of any kind planned – though in practice AMD did release a couple of drivers to fix critical security issues.
AMD’s support plans for Vega and Polaris, by contrast, still call for regular driver releases, albeit without major feature updates or performance optimizations. That means receiving bug fixes and other occasional updates as AMD sees fit to backport them to the older driver branch, but not the full scope of updates that AMD’s RDNA products are now receiving via their up-to-date mainline driver. In practice, this is much closer to how NVIDIA has handled their legacy GPU products, which have traditionally received security fixes for a minimum length of time – making for a more welcome offramp for going from fully supported to unsupported.
But regardless of what AMD is calling their driver support policy for Polaris and Vega, the end result is that major driver development for these GPU architectures has effectively come to an end, and these parts have now entered an extended support phase. AMD considers the drivers for the hardware mature, and with games increasingly taking advantage of features not supported by the hardware (e.g. Alan Wake II and mesh shaders), there’s clearly less of a need to add support/optimizations for new games to drivers for old hardware.
More broadly speaking, with the current high-end game consoles using a RDNA2-ish (DX feature level 12_2) architecture, we’re finally approaching a rather hard switch that will be leaving pre-12_2 GPUs behind. Consequently, I’m not surprised to see AMD semi-retire both Polaris and Vega at the same time – there is a clear gulf in archtiecture between GCN and the greatly reworked RDNA that underpins AMD’s more recent cards.
| AMD Recent GPU Driver Releases | |||
| Adrenaline | RDNA | GCN 4 & 5 | |
| 23.11.1 (November) | 23.20.23.01 | 23.19.05.01 | |
| 23.10.2 (October) | 23.20.17.03 | N/A | |
| 23.9.3 (September) | 23.20.11.04 | 23.19.02 | |
| 23.8.2 (August) | 23.10.29.06 | ||
As AMD enacted this driver split back in September, we can already see some of the ramifications of this with AMD’s latest drivers. Polaris and Vega did not receive an October driver release (Adrenaline 23.10), and the November release (Adrenaline 23.11.1) contains only a handful of fixes for GCN cards, as opposed to the much more extensive list of fixes and new game support for RDNA cards.
Ultimately, while the remaining GCN GPUs haven’t been put out to pasture quite yet, this is clearly the beginning of the end for a line of GPU architectures that stretches back to AMD’s 2011 GPU architecture modernization. AMD has been selling Polaris (GCN 4) cards since mid-2016 – starting over seven years ago – and in practice the core compute and graphics architecture of GCN 4 is virtually identical to the even older GCN 3 architecture. Consequently, AMD has essentially been supporting that core GPU architecture for almost 9 years at this point.
| GCN 4 & 5 Products | |||
| Desktop | Mobile | ||
| Radeon VII | Radeon 600 Mobile Series | ||
| Radeon RX Vega Series | Radeon 500 Mobile Series | ||
| Radeon Pro Duo | Radeon 400 Mobile Series | ||
| Radeon 600 Series | Ryzen Mobile 7030U Series | ||
| Radeon RX 500 Series | Ryzen Mobile 5000 Series | ||
| Radeon RX 400 Series | Ryzen Mobile 4000 Series | ||
| Ryzen 5000G Series APUs | Ryzen Mobile 3000 Series | ||
| Ryzen 4000G Series APUs | Ryzen Mobile 2000 Series | ||
| Ryzen 3000G Series APUs | |||
| Ryzen 2000G Series APUs | |||
Meanwhile, things are a little more short-lived for the newer Vega GPU architecture (GCN 5). While AMD introduced the first discrete Vega GPUs and video cards in mid-2017 – and replaced the whole lot of them in mid-2019 – that GPU architecture remained in use in current-generation products for much longer as an integrated graphics solution. AMD’s current desktop APUs, the Ryzen 5000G series, integrate a Vega GPU. And the same silicon is still sold in the mobile space as the budget-minded Ryzen Mobile 7030 series. So although the Vega architecture is only a year younger than Polaris, it has stuck around for much longer overall.
Unfortunately, that does also mean that these Vega-based APUs are also getting something of the short end of the stick when it comes to driver support, receiving only a few years of mainstream driver support before being deprioritized. Though as these are also the weakest Vega GPUs, they’re admittedly also the least likely to be used with new games. More critical here will be how long AMD supplies security fixes for the Vega GPU architecture, especially since GPU drivers are popular targets for privilege escalation attacks.
In any case, while this isn’t a eulogy for the final members of the Graphics Core Next GPU architecture – at least, not quite yet – it’s clear that, 12 years later, GCN’s time is finally approaching its end.
When Intel unveiled its range of Arc A-series desktop graphics cards last year, it introduced four models: the Arc A770, Arc A750, Arc A580, and Arc A380. However, the Arc A580, which uses a cut-down ACM-G10 GPU, never reached the market for reasons that remain unclear. On Tuesday Intel finally fleshed out the Arc desktop lineup with a 500 series card, formally and immediately launching the Arc A580 graphics card.
Intel's Arc A580 is based on the Alchemist ACM-G10 graphics processor with 3072 stream processors and that is paired with 8 GB of memory using a 256-bit interface. While the the cut-down GPU has fewer SPs than its higher-performing counterparts, it retains all of the features that the Alchemist architecture has to offer, including world-class media playback capabilities, including hardware accelerated decoding and encoding in AV1, H.264, and H.265 formats.
The card sits under the Arc A770 and Arc A750 in terms of performance, but above the Arc A380, thus targeting gamers in budget. Intel itself positions its Arc A580 for 1080p gaming against AMD's Radeon RX 6600 and NVIDIA's GeForce RTX 3050 graphics cards that have been available on the market for about two-and-a-half years.
When compared to its rivals, the Arc A580 has higher compute performance (10.445 FP32 TFLOPS vs. Radeon RX 6600's 9 FP32 TFLOPS and GeForce RTX 3050's 8 FP32 TFLOPS) as well as dramatically higher memory bandwidth (512 GB/s vs. 224 GB/s). Though as FLOPS are not everything, we'll have to see how benchmarks play out. The biggest advantage for Intel right now is going to be memory bandwidth, as Intel is shipping a card with a far wider memory bus than anything else in this class – something that AMD and NVIDIA shied away from after multiple cryptocurrency rushes and crashes.
But Intel's Arc A580 is more power hungry than its rivals: as this part is based on Intel's top-tier ACM-G10 GPU, it has the power consumption to match, with a total graphics power rating of 185W. Conversely, AMD's Radeon RX 6600 and NVIDIA's GeForce RTX 3050 are rated for 132W and 130W, respectively.
Graphics cards based on the Intel Arc A580 GPU are set to be offered by ASRock, Gunnir, and Sparkle, starting at $179. At $179, the boards are cheaper than AMD's Radeon RX 6600 ($199) and Nvidia's GeForce RTX 3050 ($199), which makes it quite a competitive offering. Meanwhile, Intel's higher-performing Arc A750 can now be obtained for $189 - $199, which somewhat reduces appeal of the new board – though it remains to be seen if those A750 prices will last.
When Asus teased its ROG Matrix GeForce RTX 4090 graphics card back at Computex, it was clear that the company's ambitions were to develop no less than the world's fastest graphics card. The company meticulously described the card's advanced printed circuit board design, voltage regulating module, and cooling system, but it never revealed two important details: actual clocks and price. This week it disclosed both: the board will clock the GPU at 2.70 GHz out-of-box and will cost $3,199, twice the price of a reference GeForce RTX 4090.
Asus proudly states that the ROG Matrix GeForce RTX 4090 is ideal for overclocking enthusiasts. The board used the AD102 GPU equipped with 16,384 CUDA cores that has a peak frequency of 2700 MHz, surpassing NVIDIA's reference boost clock of 2520 MHz. In a physically unmodified (but LN cooled) state, an extreme overclocked ROG Matrix GeForce RTX 4090 surpassed the 4 GHz GPU clock threshold earlier this year, an achievement that underscores its potential for overclocking.
Since its debut at Computex, the card has secured three World Records and five top spots, totaling seven overclocking achievements in various benchmarks, Asus says.
NVIDIA has dozens of add-in-board (AIB) partners producing factory overclocked graphics cards. But with EVGA and its Kingpin-edition graphics cards gone, there are not so many brands left which cater to demands of extreme enthusiasts. Asus is certainly one of them and with its range-topping ROG Matrix RTX 4090, the company went above and beyond with enhancements beyond reference designs.
The card employs a custom circuit board featuring a 24-phase VRM and a 12VHPWR connector, ensuring up to 600W of power for the GPU. This board is equipped with multiple sensors to oversee temperatures of various components (and even create a temperature map) and even measure currents on the card's 12VHPWR connector (more on this later).
The ROG Matrix GeForce RTX 4090 comes with a comprehensive closed-loop hybrid liquid cooling solution with a 360-mm radiator, magnetically connected fans, and RGB illumination. In a bid to improve efficiency of the cooler, Asus used a liquid metal thermal compound, which it uses for its gaming laptops and which is particularly hard to use for desktop PC components (marking a first in the GPU industry for Asus) since they tend to be located under a different angle.
The ROG Matrix RTX 4090's strengths are not solely in its hardware though. Asus has enhanced its GPU Tweak III software, adding more monitoring and overclocking capabilities that leverage the card's advanced features and sensors. Users can customize various settings, including power targets, GPU voltage, and fan speed. The software also offers real-time temperature insights and tracks the card's performance at varying power settings.
Another notable aspect is the card's Power Detector+ feature. This function examines the 12VHPWR connector, monitoring currents across all power rails to identify any irregularities, then recommends customers to reconnect the notorious plug if needed.
Meanwhile, performance of the ROG Matrix RTX 4090 comes at a cost as the product's price doubles that of a standard GeForce RTX 4090. This greatly devalues the product in the eyes of average people. But the Asus ROG Matrix RTX 4090 is a niche product. It targets hardcore overclocking enthusiasts eager to maximize their hardware's performance. This card is for those who relish fine-tuning their systems for minor benchmarking improvements, making it a trophy piece for tech enthusiasts.
ASRock this week has introduced an Intel Arc-based A310 Low Profile 4 GB graphics card. Aimed at the lower power/size/cost portion of the video card market, the entry-level A310 card carries a cut-down ACM-G11 GPU with 768 stream processors operating at 2.0 GHz and a 64-bit memory interface that connects 4 GB of GDDR6. And while its performance isn't anything to write home about, it fills out ASRock's lineup with a budget video card that can also easily be placed into low profile systems.
Performance vise, ASRock's Arc A310 Low Profile graphics card should be comparable to higher-end integrated graphics solutions - which is to say that it's more for general desktop work than demanding games. Notably, as a current-generation Arc card, it incorporates Intel's feature-packed media playback engine, with hardware encoding and decoding for the majority of popular formats, such as AV1, H.264, H.265, and VP9, which makes it a good solution for home theater PCs (HTPCs) as well as office PCs.
The low-profile card offers two display outputs, a DisplayPort (2.0) and an HDMI port (2.0b). In terms of power consumption, this is a sub-75W card (ASRock doesn't give a more precise figure than that), so you won't find any external power connectors as it's entirely bus-powered. Meanwhile, it looks like ASRock has opted to recycle the same dual-slot dual-fan cooling system they use for their other low-profile Arc card, the low-profile ASRock Arc A380.
Wrapping things up, ASRock has not listed a price for the card. Though judging from the cost of other Arc A310 cards on the market, it's likely to end up hitting shelves in the low $100 range.
Modern graphics cards are notorious for their massive cooling systems and dimensions that are hard to fit into mainstream PC chassis and almost impossible to fit into compact PC cases that are rather popular among gamers. MSI this week attempted to address these concerns by announcing a family of GeForce RTX 40-series graphics cards called Gaming Slim that are meant to be slimmer than a typical add-in-board. But are they?
Judging by MSI's GeForce RTX 4070 Ti Gaming X Slim (which measures 307×125×51 mm and weights 1086 grams), it is indeed considerably more compact than MSI's GeForce RTX 4070 Ti Gaming X Trio (337×140×62 mm @ 1594 grams). What is no less important is that the Gaming Slim is nearly as fast as its full-fat counterpart with its maximum 2745 MHz GPU clock vs. 2760 MHz.
With that said, despite the "slim" designation, we're still looking at a graphics card that is 2.5 slots (51 mm) wide. So the most densely packed of systems that can't take a card over 2 slots will still need to look elsewhere. For everything else, it will certainly be easier to assemble a PC with MSI's Gaming Slim graphics card due to its squeezed dimensions.
MSI says that its Gaming Slim-badged graphics cards will provide signature features of its Gaming-branded graphics cards, including factory overclocked GPU and enhanced cooling system, but will be thinner and lighter, enabling 'flexible system assembly.' MSI further noted that it would offer all of its Gaming GeForce RTX 40-series products in Gaming Slim flavor, including GeForce RTX 4090, RTX 4080, RTX 4070, and RTX 4060.
MSI says that its Gaming Slim graphics cards will come in black and white versions, will use the company's latest cooling solutions (Tri Frozr triple-fan cooling system with Fan Torx Fan 5.0 fans and up to eight pipes), and addressable RGB LEDs.
For now, the company has Gaming Slim versions of GeForce RTX 4070 Ti, GeForce RTX 4070, and GeForce RTX 4060 Ti, so the higher-end products will probably be available at a later date. Meanwhile, we can only wonder how compact will MSI manage to get NVIDIA's GeForce RTX 4090 and whether the company's GeForce RTX 4090 Gaming Slim will be more compact than NVIDIA's GeForce RTX 4090 Founders Edition.
Alongside this morning’s announcement of the new Radeon RX 7800 XT and RX 7700 XT video cards, AMD is also using their Gamescom launch event to deliver an update on the state of their Radeon software stack. The company has been working on a couple of performance-improvement projects since the launch of the Radeon RX 7000 series in late 2022, including the highly anticipated FSR 3, and they’re finally offering a brief update on those projects ahead of their September launches.
First and foremost, AMD is offering a bit of a teaser update on Fidelity FX Super Resolution 3 (FSR 3), their frame interpolation (frame generation) technology that is the company’s answer to NVIDIA’s DLSS 3 frame generation feature. First announced back at the Radeon RX 7900 series launch in 2022, at the time AMD only offered the broadest of details on the technology, with the unspoken implication being that they had only recently started development.
At a high level, FSR 3 was (or rather, will be) AMD’s open source take on frame interpolation, similar to FSR 2 and the rest of the Fidelity FX suite of game effects. With FSR 3, AMD is putting together a portable frame interpolation technique that they’re dubbing “AMD Fluid Motion Frames” and, unlike DLSS, is not vendor proprietary and can work on a variety of cards from multiple vendors. Furthermore, the source code for FSR3 will be freely available as part of AMD’s GPUOpen community.
Since that initial announcement, AMD has not had anything else of substance to say on the state of FSR 3 development. But at last, the first shipping version of FSR 3 is in sight, with AMD expecting to bring it to the first two games next month.
Ahead of that launch, they are offering a small taste of what’s to come, with some benchmark numbers for the technology in action on Forspoken. Using a combination of Fluid Motion Frames, Anti-Lag+, and temporal image upscaling, AMD was able to boost 4K performance on Forspoken from 36fps to 122fps.
Notably, AMD is using “Performance” mode here, which for temporal upscaling means rendering at one-quarter the desired resolution of a game – in this case, rendering at 1080p for a 4K/2160p output. So a good deal of the heavy lifting is being done by temporal upscaling, but not all of it.
AMD has also published a set of numbers without temporal upscaling, using their new native anti-aliasing mode, which renders at the desired output resolution and then uses temporal techniques for AA, and combines that with Fluid Motion Frames. In that case, performance at 1440p goes from 64fps to 106fps.
For the time being, these are the only two sets of data points AMD is providing. Otherwise, the screenshots included in their press deck are not nearly high enough in quality to make any kind of meaningful image quality comparisons, and AMD hasn’t published any videos of the technology in action. So convincing visual evidence of FSR 3 in action is, at least ahead of today’s big reveal, lacking. But it is a start none the less.
As for the technical underpinnings, AMD has answered a few questions relating to FSR3/Fluid Motion Frames, but the company is not doing a deep dive on the technology at this time. So there remains a litany of unanswered questions about its implementation.
With regards to compatibility, according to AMD FSR3 will work on any RDNA (1) architecture GPU or newer, or equivalent hardware. RDNA (1) is a DirectX Feature Level 12_1 architecture, which means equivalent hardware spans a pretty wide gamut of hardware, potentially going back as far as NVIDIA’s Maxwell 2 (GTX 900) architecture. That said, I suspect there’s more to compatibility than just DirectX feature levels, but AMD isn’t saying much more about system requirements. What they are saying, at least, is that while it will work on RDNA (1) architecture GPUs, they recommend RDNA 2/RDNA 3 products for the best performance.
Along with targeting a wide range of PC video cards, AMD is also explicitly noting that they’re targeting game consoles as well. So in the future, game developers will be able to integrate FSR3 and have it interpolate frames on the PlayStation 5 and Xbox Series X|S consoles, both of which are based on AMD RDNA 2 architecture GPUs.
Underpinning FSR 3’s operation, in our briefing AMD made it clear that it would require motion vectors, similar to FSR 2’s temporal upscaling, as well as rival NVIDIA’s DLSS 3 interpolation. The use of motion vectors is a big part of why FSR 3 requires per-game integration – and a big part of delivering high quality interpolated frames. Throughout our call “optical flow” did not come up, but, frankly, it’s hard to envision AMD not making use of optical flow fields as well as part of their implementation. In which case they may be relying on D3D12 motion estimation as a generic baseline implementation, since it wouldn’t require accessing each vendor’s proprietary integrated optical flow engine.
What’s not necessary, however, is AI/machine learning hardware. Since AMD is targeting the consoles, they wouldn’t be able to rely on it anyhow.
AMD also says that FSR 3 includes further latency reduction technologies (which are needed to hide the latency of interpolation). It’s unclear if this is something equivalent to NVIDIA’s Reflex marker system, or if it’s something else entirely.
The first two games to get FSR 3 support will be the aforementioned Forspoken, as well as the recently launched Immortals of Aveum. AMD expects FSR 3 to be patched in to both games here in September – presumably towards the later part of the month.
Looking farther forward, AMD has lined up several other developers and games to support the technology, including RPG-turned-technology-showcase Cyperpunk 2077. Given that Forspoken and Immortals of Aveum are essentially going to be the prototypes for testing FSR 3, AMD isn’t saying when support for the technology will come to these other games. Though with plans to make it available soon as an Unreal Engine plugin, the company certainly has its eyes on enabling wide-scale deployment of the technology over time.
For now, this is only a brief teaser of details. I expect that AMD will have a lot more to say about FSR 3, and to disclose, once the FSR 3 patches for Forspoken and Immortals of Aveum are ready to be released.
Second up, we have Hypr-RX. AMD’s smorgasbord feature combines the company’s Radeon Super Resolution (spatial upscaling), Radeon Anti-Lag+ (frame queue management), and Radeon Boost (dynamic resolution scaling). All three technologies are already available in AMD’s drivers today, however they can’t all currently be used together. Super Resolution and Boost both touch the rendering resolution of a game, and Anti-Lag steps on the toes of Boost’s dynamic resolution adjustments.
Hypr-RX, in turn, is designed to bring all three technologies together to make them compatible with one another, and to put the collective set of features behind a single toggle. In short, if you turn on Hypr-RX, AMD’s drivers will use all of the tricks available to improve game performance.
Hypr-RX was supposed to launch by the end of the first half of the year, a date that has since come and gone. But, although a few months late, AMD has finally finished pulling together the feature, just in time for the Radeon RX 7800 XT launch.
To that end, AMD will be shipping Hypr-RX in their next Radeon driver update, which is due on September 6th. This will be the launch driver set posted for the RX 7800 XT, bringing support for the new card and AMD’s newest software features all at once. On that note, it bears mentioning that Hypr-RX requires an RDNA 3 GPU, meaning it’s only available for Radeon RX 7000 video cards as well as the Ryzen Mobile 7040HS CPU family.
Taking a quick look at performance, AMD has released some benchmark numbers showing both the latency and frame rates for several games on the RX 7800 XT. In all cases latency is down and framerates are up, varying on a game-by-game basis. As these individual technologies are already available today, there’s not much new to say about them, but given the overlap in their abilities and the resulting technical hurdles, it’s good to see that AMD finally has them playing nicely together.
But with FSR 3 and its frame interpolation abilities soon to become available, AMD won’t be stopping there for Hypr-RX. The next item on AMD’s to-do list is to add Fluid Motion Frame support to Hypr-RX, allowing AMD’s drivers to use frame interpolation (frame generation) to further improve game performance.
This is a bigger and more interesting challenge than it may first appear, because AMD is essentially promising to (try to) bring frame interpolation to every game on a driver level. FSR 3 requires that the technology be built in to each individual game, in part because it relies on motion vector data. That motion vector data is not available to driver-level overrides, which is why Hypr-RX’s Radeon Super Resolution ability is only a spatial upscaling technology.
Put another way: AMD apparently thinks they can do frame interpolation without motion vectors, and still achieve a good enough degree of image quality. It’s a rather audacious goal, and it will be interesting to see how it turns out in the future.
Wrapping things up, AMD’s current implementation of Hypr-RX will be available on September 6th as part of their new driver package. Meanwhile Hypr-RX with frame interpolation is a work in progress, and will be coming at a later date.
With the Gamescom convention taking place in Germany this week, AMD is using Europe’s largest video game trade show as the venue for their latest Radeon video card announcements. This morning the company is announcing their long-awaited middle members of the Radeon RX 7000 series, the Radeon RX 7800 XT and Radeon RX 7700 XT. Aimed at the 1440p gaming market and based on AMD’s new Navi 32 GPU, the new cards are designed to slot in to the middle of AMD’s product stack, offering a set of potent RDNA 3 architecture video cards for gamers who don’t need the bleeding-edge performance (and wallet-bleeding costs) of the Radeon RX 7900 series cards.
With today’s announcement being just that – an announcement – the retail launch of these cards will follow in a couple of weeks, on Wednesday, September 6th. That date also not-so-coincidentally happens to be the release date for Bethesda’s ARPG Starfield, for which AMD is the game’s exclusive PC hardware partner, and which AMD will be offering with the new Radeon cards as part of the company’s latest game bundle. So for AMD, the stars are aligning to make September 6th a big day for the company’s GPU division.
But as for us, we’re here to talk about hardware, so let’s discuss the Radeon RX 7800 XT and 7700 XT, as well as the Navi 32 GPU that underpins them.
Earlier this month, AMD said that it would refresh its lineup of Radeon RX graphics cards for gamers this quarter, and apparently, the new products are set to be announced next week at Gamescom in Cologne, Germany.
"Please join the @AMDRadeon team at Gamescom next week for our next major product announcements," said Scott Herkelman, senior vice president and general manager graphics business unit at AMD, in an X post.
Frank Azor, chief architect of gaming solutions and marketing at AMD, linked Herkelman's post and reaffirmed that the company has 'some news coming next week.'
Indeed, Gamescom is a good place to announce new products. The trade show traditionally attracts hundreds of thousands of visitors, making it one of the biggest gaming events in the world. The fair is aimed at the general public and industry professionals. The event is split between an Entertainment Area for the public, where fans can try out upcoming games, and a Business Area for trade visitors to communicate and conduct deals. AMD plans to hold its AMD Gaming Festival 2023 on Friday, August 25, at Hall 7, which starts at 12:00.
While AMD does not disclose what exactly it plans to announce at Gamescom, it is about time for AMD to fill a gap in its Radeon RX 7000-series product line that spans from the Radeon RX 7600 ($270) and goes all the way to the Radeon RX 7900 XT ($900). This void is currently being filled by the older Radeon RX 6000-series and the Radeon RX 7900 GRE ($650), which is hard to get. Essentially, AMD does not have a direct answer to NVIDIA's reasonably popular GeForce RTX 4070 at the moment.
It's speculated that AMD's next move is to unveil the Navi 32, a GPU from its RDNA 3-based lineup which would position itself between the existing Navi 31 and Navi 33 GPUs. AMD's Navi 32 is anticipated to be the foundation for the Radeon RX 7700 and Radeon RX 7800 series, which will compete against Nvidia's performance mainstream and higher-end GeForce RTX 40-series products.
When NVIDIA began to roll out their Ada Lovelace architecture to the workstation market, the company introduced its new flagship RTX 6000 Ada graphics card meant to offer the highest performance possible as well as its quite spectacular RTX 4000 SFF board that delivers formidable performance in a tiny package. The gap between the two solutions is vast, and on Tuesday, the company finally unveiled new products that fill it.
NVIDIA's Ada Lovelace-based RTX-series professional graphics cards — the workstation-oriented RTX 4000 20GB, RTX 4500 24GB, RTX 5000 32GB, and the datacenter-bound L40S — graphics are designed for demanding graphics and artificial intelligence workloads, such as computer-aided design, digital content creation, real-time rendering, and basic simulations that are fine with FP32 precision. The new graphics solutions complement NVIDIA's Ada Lovelace-based workstation boards that have been announced: the midrange RTX 4000 SFF and the ultra-high-end RTX 6000 Ada. Meanwhile, NVIDIA's previous-generation offerings will continue to serve entry-level workstations based on its Ampere and Turing architectures.
Now, let us cover the new graphics boards in more detail.
NVIDIA's RTX 4000 20GB is powered by the AD104 graphics processor with 6,144 CUDA cores that promises a peak performance of 26.7 FP32 TFLOPS, which is considerably higher than 19.2 FP32 TFLOPS delivered by the RTX 4000 SFF that features the same GPU with the same configuration albeit working at lower clocks and therefore consuming up to 130W. Unlike the small form-factor board, this uses a full-height PCB but a single-slot cooling system. The novelty is slated for a September release with an MSRP of $1,250.
The more powerful Ada Lovelace-based workstation board is called the RTX 4500, and it uses the AD104 GPU with 7,680 CUDA cores to deliver a compute performance of up to 39.6 FP32 TFLOPS at up to 210W. The board employs a dual-slot cooling system and will be available for $2,250 sometime in October.
Finally, NVIDIA is introducing its RTX 5000 professional graphics card that utilizes the AD102 graphics processor with 12,800 CUDA cores (i.e., a very significant cut down) to achieve a compute performance of 65.3 FP32 TFLOPS at 250W. This board is set to hit the market now for $4,000, which is significantly lower compared to $6,800 for NVIDIA's flagship RTX 6000 Ada product.
| NVIDIA Ada Lovelace Professional Graphics Cards | |||||||
| RTX 4000 SFF | RTX 4000 | RTX 4500 | RTX 5000 | RTX 6000 | L40S Ada | ||
| GPU | AD104 | AD104 | AD104 | AD102 | AD102 | AD102 | |
| CUDA Cores | 6,144 | 6,144 | 7,680 | 1,2800 | 1,8176 | 1,8176 | |
| Memory | 20 GB | 24 GB | 32 GB | 48 GB | |||
| Power | 70W | 130W | 210W | 250W | 300W | ? | |
| Cooling | dual-slot, blower | single-slot, blower | dual-slot, blower | passive | |||
| MSRP | $1,250 | $1,250 | $2,250 | $4,000 | $6,800 | ? | |
NVIDIA's latest ProViz graphics boards are set to be integrated into the upcoming workstation lineups of renowned companies, including Boxx, Dell, HP, Lambda, and Lenovo. Additionally, the graphics cards will be available for purchase from select graphics card makers like Leadtek, PNY, and Ryoyo, as well as major resellers like Arrow and Ingram. Meanwhile, there will be an Ada Lovelace professional graphics board that will unlikely be available separately.
Catering to the needs of professionals using remote workstations, NVIDIA is launching the L40S Ada datacenter card. The board carries the AD102 graphics processor with 18,176 active CUDA cores, delivering a staggering 91.6 FP32 TFLOPS performance. The product is initially set for NVIDIA's OVX servers that can be used to enable cloud AI and virtual desktop infrastructure. Still, it is reasonable to expect other AI and VDI infrastructure makers to adopt the L40S Ada board. Interestingly, despite being a data center-oriented product with passive cooling, the L40S Ada includes display outputs, making it suitable for workstations given adequate airflow inside or an attached blower. NVIDIA does not publish the pricing of its OVX machine or the L40S Ada card.
"OVX systems with NVIDIA L40S GPUs accelerate AI, graphics, and video processing workloads and meet the demanding performance requirements of an ever-increasing set of complex and diverse applications," said Bob Pette, vice president of professional visualization at NVIDIA
At SIGGRAPH in Los Angeles, NVIDIA unveiled a new variant of their GH200' superchip,' which is set to be the world's first GPU chip to be equipped with HBM3e memory. Designed to crunch the world's most complex generative AI workloads, the NVIDIA GH200 platform is designed to push the envelope of accelerated computing. Pooling their strengths in both the GPU space and growing efforts in the CPU space, NVIDIA is looking to deliver a semi-integrated design to conquer the highly competitive and complicated high-performance computing (HPC) market.
Although we've covered some of the finer details of NVIDIA's Grace Hopper-related announcements, including their disclosure that GH200 has entered into full production, NVIDIA's latest announcement is a new GH200 variant with HBM3e memory is coming later, in Q2 of 2024, to be exact. This is in addition to the GH200 with HBM3 already announced and is currently in production and due to land later this year. This means NVIDIA has two versions of the same product, with GH200 incorporating HBM3 incoming and GH200 with HBM3e set to come later.
| NVIDIA Grace Hopper Specifications | ||||
| Grace Hopper (GH200) w/HBM3 | Grace Hopper (GH200) w/HBM3e | |||
| CPU Cores | 72 | 72 | ||
| CPU Architecture | Arm Neoverse V2 | Arm Neoverse V2 | ||
| CPU Memory Capacity | <=480GB LPDDR5X (ECC) | <=480GB LPDDR5X (ECC) | ||
| CPU Memory Bandwidth | <=512GB/sec | <=512GB/sec | ||
| GPU SMs | 132 | 132? | ||
| GPU Tensor Cores | 528 | 528? | ||
| GPU Architecture | Hopper | Hopper | ||
| GPU Memory Capcity | 96GB (Physical) <=96GB (Available) |
144GB (Physical) 141GB (Available) |
||
| GPU Memory Bandwidth | <=4TB/sec | 5TB/sec | ||
| GPU-to-CPU Interface | 900GB/sec NVLink 4 |
900GB/sec NVLink 4 |
||
| TDP | 450W - 1000W | 450W - 1000W | ||
| Manufacturing Process | TSMC 4N | TSMC 4N | ||
| Interface | Superchip | Superchip | ||
| Available | H2'2023 | Q2'2024 | ||
During their keynote at SIGGRAPH 2023, President and CEO of NVIDIA, Jensen Huang, said, "To meet surging demand for generative AI, data centers require accelerated computing platforms with specialized needs." Jensen also went on to say, "The new GH200 Grace Hopper Superchip platform delivers this with exceptional memory technology and bandwidth to improve throughput, the ability to connect GPUs to aggregate performance without compromise, and a server design that can be easily deployed across the entire data center."
NVIDIA's GH200 GPU is set to be the world's first chip to ship with HBM3e memory, an updated version of the high-bandwidth memory with even greater bandwidth and, critically for NVIDIA, higher capacity 24GB stacks. This will allow NVIDIA to expand its local GPU memory from 96GB per GPU to 144GB (6 x 24GB stacks), a 50% increase that should be especially welcome for the AI market, where top models are massive in size and often memory capacity bound. In a dual configuration setup, it will be available with up to 282 GB of HBM3e memory, which NVIDIA states "delivers up to 3.5 x more memory capacity and 3 x more bandwidth than the current generation offering."
Perhaps one of the most notable details NVIDIA shares is that the incoming GH200 GPU with HBM3e is 'fully' compatible with the already announced NVIDIA MGX server specification, unveiled at Computex. This allows system manufacturers to have over 100 different variations of servers that can be deployed and is designed to offer a quick and cost-effective upgrade method.
NVIDIA claims that the GH200 GPU with HBM3e provides up to 50% faster memory performance than the current HBM3 memory and delivers up to 10 TB/s of bandwidth, with up to 5 TB/s per chip.
We've already covered the announced DGX GH200 AI Supercomputer built around NVIDIA's Grace Hopper platform. The DGX GH200 is a 24-rack cluster fully built on NVIDIA's architecture, with each a single DGX GH200 combining 256 chips and offering 120 TB of CPU-attached memory. These are connected using NVIDIA's NVLink, which has up to 96 local L1 switches providing immediate and instantaneous communications between GH200 blades. NVIDIA's NVLink allows the deployments to work together with a high-speed and coherent interconnect, giving the GH200 full access to CPU memory and allowing access for up to 1.2 TB of memory when in a dual configuration.
NVIDIA states that leading system manufacturers are expected to deliver GH200-based systems with HBM3e memory sometime in Q2 of 2024. It should also be noted that GH200 with HBM3 memory is currently in full production and is set to be launched by the end of this year. We expect to hear more about GH200 with HBM3e memory from NVIDIA in the coming months
Colorful has quietly introduced GeForce RTX 4060 Ti graphics cards in Mini-ITX form-factor that combine compact dimensions and performance of Nvidia's latest Ada Lovelace GPUs. In fact, with the boosted performance of 22 FP32 TFLOPS, Colorful's iGame GeForce RTX 4060 Ti Mini OC is likely the highest-performing Mini-ITX graphics card that has been launched to date.
Looking at the options, Colorful has two iGame GeForce RTX 4060 Ti Mini graphics cards in Mini-ITX form-factor: one with 8 GB of GDDR6 memory and another with 16 GB of GDDR6 SGRAM. Both boards are naturally based on Nvidia's AD106 GPU with 4352 CUDA cores running at 2310 MHz – 2580 MHz, which is slightly higher than Nvidia's recommended 2540 MHz. The board has four display outputs (three DisplayPort, one HDMI), like fully-fledged GeForce RTX 4060 Ti boards.
Since the boards have a relatively simplistic 6+2-phase voltage regulating module (VRM), it is unlikely that the cards were designed with overclocking in mind, so it is doubtful that these devices can be overclocked significantly. Furthermore, they are equipped with a dual-slot single-fan cooling system with four heat pipes, good enough to dissipate 160W ~ 165W of heat generated by Colorful's iGame GeForce RTX 4060 Ti Mini. Still, I doubt this cooler will enable great overclocking potential.
Colorful's iGame GeForce RTX 4060 Ti Mini graphics cards are said to be 199.5 mm long, which is somewhat longer than typical Mini-ITX motherboards, so owners will probably have to ensure compatibility of these AIBs with the chassis.
Compact Mini-ITX PCs are rather popular among gamers these days, but due to the high power consumption of Nvidia's previous-generation graphics processors did not allow makers of add-in-boards to make Mini-ITX versions of their midrange products. With Ada Lovelace, Nvidia opted to reduce the power consumption of its GeForce RTX 4060-series, which enabled makers of AIBs to experiment with form factors and come up with Mini-ITX versions of GeForce RTX 4060 Ti.
Unfortunately, Colorful's graphics cards are rare guests in North American and European retailers, so those interested in the company's iGame GeForce RTX 4060 Ti Mini graphics cards should probably buy directly or from retailers like JD.com.
Given that Nvidia's GeForce RTX 4060 Ti GPU is readily available from the green company, other makers of graphics cards may follow Colorful with their Mini-ITX versions at some point.
The relatively low power consumption of Nvidia's GeForce RTX 4060 graphics processor allows graphics card makers to experiment with the form factors of their products. We have already seen Mini-ITX GeForce RTX 4060 graphics cards, and late last week, Gigabyte introduced a low-profile GeForce RTX 4060 that can fit into miniature desktops and provide decent performance in games.
The Gigabyte GeForce RTX 4060 OC Low Profile 8G is based on NVIDIA's AD107 GPU with 3072 CUDA cores that are paired with 8 GB of 17 GT/s GDDR6 memory using a 128-but interface. To justify the OC (overclocked) moniker in the product name, Gigabyte even clocked the graphics processor at 2475 MHz, which is 15 MHz higher than Nvidia's recommendations for the RTX 4060 model.
Using a graphics board that requires an eight-pin auxiliary PCIe power connector, it is equipped with a dual-slot triple-fan cooling system featuring dozens of thin aluminum. We can only guess whether the cooler is quiet and whether it is good enough to enable further overclocking, but at least Gigabyte guarantees a GPU boost clock of up to 2475 MHz.
Touching more on the cooler, it is longer than the printed circuit board itself, making the graphics card 182 mm long, so owners of compact systems should measure their chassis to ensure compatibility. Most low-profile PC cases are pretty long, but there are also tiny chassis that may be too small for this card.
Despite being low profile, Gigabyte's GV-N4060OC-8GL has four display outputs: two DisplayPort 1.4a (up to 4Kp120 or up to 5Kp60) and two HDMI 2.1a (up to 5Kp60 or up to 8Kp60 with DSC), so it can be used for rather serious PCs with up to four monitors.
Gigabyte has not disclosed the recommended pricing of their GeForce RTX 4060 OC Low Profile 8G graphics card. Considering that prices of most GeForce RTX products are hovering around recommended $299 price point, it is unlikely that Gigabyte will attempt to charge a huge premium for the unique form factor of its low-profile GeForce RTX 4060. Still, the compact dimensions are undoubtedly a significant differentiator of this product, and GIGABYTE will likely try to earn something extra from it.
Source: Gigabyte
As part of their quarterly earnings call this week, AMD revealed that the company is getting ready to launch new enthusiast-class Radeon RX 7000-series graphics cards in the coming months. To date, the company has launched cards for the top and bottom portions of their product stack, leaving a noticeable gap for higher performing cards that the company needs to fill to fully flesh out the current card lineup.
"We are on track to further expand our RDNA 3 GPU offerings with the launch of new, enthusiast-class Radeon 7000 series cards in the third quarter," said Lisa Su, chief executive of AMD, at the company's earnings call with analysts and investors.
So far, AMD has introduced four RDNA 3-based Radeon RX 7000-series desktop graphics cards aimed at diversified market segments: three Radeon RX 7900-series offerings for enthusiasts who can spend between $650 and $1000 on a graphics card, and the Radeon RX 7600 product for mainstream gamers at roughly $270. This has left an empty space for higher performing cards for cost-conscientious enthusiasts that, for the moment, is being met by NVIDIA's GeForce RTX 4000-series as well as previous-generation Radeon RX 6000-series boards. In particular, AMD currently lacks something current to compete with NVIDIA's modestly well received GeForce RTX 4070.
AMD is believed to have only one GPU left in its Navi 30 range, Navi 32, which would slot in between the current Navi 31 and Navi 33 parts. Navi 32, in turn, is expected to power both Radeon RX 7700 and RX 7800 product families. That said, one thing that remains to be seen is whether the company will decide to go after volume first this quarter and start things off with the RX 7700 series, or after higher margins and reveal its Radeon RX 7800 series first.
AMD's gaming segment revenue was $1.6 billion in Q2 2023, down 4% year-over-year and 10% sequentially primarily due to lower sales of gaming graphics cards. Unit sales of graphics processors in Q2 are typically lower than their shipments in Q1, so a 10% quarter-over-quarter decrease is not surprising. Meanwhile, a 4% drop YoY indicates that appeal of AMD's discrete GPUs was lower in Q2 2023 compared to Q2 2022, an indicator that the company needs new products.
As AMD continues to launch their full graphics product stacks based on their latest RDNA 3 architecture GPUs, the company is now preparing their next wave of professional cards under the Radeon Pro lineup. Following the launch of their high-end Radeon Pro W7900 and W7800 graphics cards back in the second quarter of this year, today the company is announcing the low-to-mid-range members of the Radeon Pro W7000 series: the Radeon Pro W7500 and Radeon Pro W7600. Both based on AMD’s monolithic Navi 33 silicon, the latest Radeon Pro parts will hit the shelves a bit later this quarter.
The two cards, as a whole, will make up what AMD defines as the mid-range segment of their professional video card market. And like their flagship counterparts, AMD is counting on a combination of RDNA 3’s advanced features, including AV1 encoding support, improved compute and ray tracing throughput, and DisplayPort 2.1 outputs to help drive sales of the new video cards. That, and as is tradition, significantly undercutting NVIDIA’s competing professional cards.
Intel has quietly released two new Arc Alchemist-series graphics processors for laptops. The new Arc A530M and Arc A570M target mid-range notebooks designed for light gaming. Perhaps the most intriguing thing about the new mobile GPUs is that they use previously unreleased ACM-G12 silicon.
Intel's Arc A530M GPU comes with 12 Xe cores and 1536 stream processors operating at 1300 MHz, which clearly distinguishes it from the company's entry-level Arc A370M GPU that only has eight Xe cores and 1024 stream processors. Meanwile, the Arc A570M features 16 Xe cores and 2048 stream processors running at 1300 MHz, which makes it clearly faster than the previously released Arc A550M with the same number of SPs at 900 MHz, but does not allow it to challenge the Arc A730M that has 3072 SPs working at 1100 MHz.
One interesting wrinkle about the Arc A530M and Arc A570M is that they seem to be based on Intel's yet-to-be-confirmed ACM-G12 GPU, according to Bionic_Squash. This graphics processor reportedly has 16 Xe clusters, which means that it sits above the ACM-G11 with eight Xe clusters and Arc-G10 with 16 Xe clusters in total. Intel yet has to formally confirm that it uses its unannounced ACM-G12 silicon for the A530M and A570M parts.
| Intel Arc Comparison | ||||||
| Arc A370M | Arc A530M | Arc A550M | Arc A570M | Arc A730M |
||
| Stream Processors | 1024 | 1536 | 2048 | 2048 | 3072 | |
| Xe-cores | 8 | 12 | 16 | 16 | 24 | |
| Render Slices | 2 | 3 | 4 | 4 | 6 | |
| Ray Tracing Units | 8 | 12 | 16 | 16 | 24 | |
| Xe Matrix Extensions (XMX) Engines | 128 | 192 | 256 | 256 | 384 | |
| Xe Vector Engines | 128 | 192 | 256 | 256 | 384 | |
| Graphics Clock | 1550 MHz | 1300 MHz | 900 MHz | 1300 MHz | 1100 MHz | |
| TGP | 35-50W | 65W-95W | 60W | 75W-95W | 80W-120W | |
| PCI Express | PCIe 4.0 x8 | PCIe 4.0 x16 | ||||
| Memory Size | 4 GB | 4 GB 8 GB |
8 GB | 8 GB | 12 GB | |
| Memory Type | GDDR6 | |||||
| Graphics Memory Interface | 64 bit | ? | 128 bit | ? | 192 bit | |
| Graphics Memory Bandwidth | 112 GB/s | ? | 224 GB/s | ? | 336 GB/s | |
| Graphics Memory Speed | 14 Gbps | ? | 14 Gbps | ? | 14 Gbps | |
One of the things that strikes the eye about the new mobile GPUs is their thermal graphics power between 65W and 95W for the Arc A530M as well as between 75W and 95W for the Arc A570M. By contrast, the Arc A550M is rated for a 60W TGP, which makes it a considerably better choice than the Arc A530M both from performance and from battery life point of view.
What remains to be seen is whether Intel uses its ACM-G12 graphics processor for desktop parts too. While the company has formally announced its Arc A580 with 3072 stream processors, this part was based on the ACM-G10 and never came to market possibly because Intel did not want to address entry-level gaming market segment. It is unclear whether Intel is interested in rolling out a discrete desktop offering that would be positioned even below the unreleased Arc A580.
Intel's newly released Arc A530M and Arc A570M are already supported by Intel's latest graphics drivers.
Sources: Intel Ark (1, 2), Bionic_Squash
One of design perks of NVIDIA's GeForce RTX 4060 is its relatively low power consumption, which has allowed graphics cards makers to produce compact add-in-boards without the massive heatsink and VRM setups required by higher-end cards. At 115 Watts, the power consumption of the card is low enough that it makes even Mini-ITX-sized cards practical, which is great news for the SFF PC community. And to that end, we've seen a few manufacturers introduce Mini-ITX GeForce RTX 4060 designs, with Lenovo now joining them as the latest member of the club with their own Mini-ITX 4060 card.
Specifications-wise, Lenovo's Mini-ITX GeForce RTX 4060 is exactly what you'd expect: it a stock-configuration RTX 4060 that carries NVIDIA's AD107 graphics processor with 3072 CUDA cores that is mated to 8 GB of GDDR6 memory using a 128-but interface. Lenovo does not disclose display outputs configuration (though expect both DisplayPort and HDMI to be present), though we see an eight-pin auxiliary PCIe power connector on the back of the board.
Meanwhile, the video card is equipped with a dual-slot, single-fan cooling system, which is typical for cards in this segment. The small heatsink and single fan is a big part in what's enabled Lenovo to build such a small card, ensuring it will fit in a Mini-ITX system (at least so long as it can accept dual-slot cards).
Overall, with its size and scale as the world's largest PC maker, we've seen Lenovo designs and manufacture scores of components over the years for their own internal use, with their Mini-ITX video card being the latest example. For the moment, Lenovo's Mini-ITX GeForce RTX 4060 is exclusively used for the company's IdeaCentre GeekPro 2023 system that is sold on JD.com. The PC is powered by Intel's Core i5-13400F and is outfitted with 16GB of RAM along with a 1TB SSD.
And while Lenovo is only using the card internally for now, there's also a chance the card could eventually make it to retail as a grey market product. The large scale of the company that makes internal component production viable also means that Lenovo parts sometimes end up on etailer shelves, especially if the company has excess stock to move.
As Intel's slate of video card board partners has expanded over the last year, so has the variety in the ecosystem, as the newly minted partners set out to design their own unique products around Intel's fledgling GPU family. This, thankfully, is starting to include some underserved markets such as low profile vide cards, which have been left behind in the focus on bigger-and-better video cards. Which brings us to ASRock's latest Arc A380 graphics card, a new low-profile A380 design that brings Intel's entry-level discrete GPU to smaller systems.
As far as specifications are concerned, ASRock's Arc A380 Low Profile 6GB (A380 LP 6G) is a typical Arc A380 product that carries Intel's ACM-G11 GPU (1024 stream processors, 2.0 GHz) coupled with 6GB of GDDR6 memory attached via a 96-bit interface. Remarkably, the card maintains a TDP below 75W, which eliminates the need for an extra power connector. This makes it a potential upgrade for small form factor machines and older PCs that do not have a spare PCIe power connector.
It should be noted, however, that while the A380 LP 6G is a low-profile card, it is still a relatively powerful card, necessitating a dual-slot cooler. So while the card should work with the majority of compact PCs, it may still be a bit too big for the smallest of systems.
The low-profile design of the ASRock Arc A380 graphics card suggests that it is not necessarily targeting even entry-level gaming machines. Instead, it appears to be a reasonable choice for minimalist office PCs and home theater PCs (HTPCs). Speaking of office PCs, ASRock's Arc A380 Low Profile 6GB graphics card only has two display outputs: one DisplayPort 2.0 and one HDMI 2.0b. This characteristic sets a constraint for applications that require more than two monitors. Of course, far not all office PCs need three or more displays, but ASRock's board will be unusable for things like video walls that are typically driven by compact PCs.
ASRock's low-profile Arc card should be available at retail soon. Though at least for the moment, the company is not listing a price for the pint-sized product.
AMD this morning is formally announcing the launch of the latest version of its GPU compute software stack, ROCm 5.7. Along with making several important updates to the software stack itself – particularly around improving support for large language models (LLMs) and other machine learning toolkits – the company has also published a blog post outlining the future hardware development plans for the stack. In short, the company will be bringing official support to a limited set of RDNA 3 architecture video cards starting this fall.
AMD’s counterpart to NVIDIA’s CUDA and Intel’s OneAPI software stacks, AMD has historically pursued a narrower hardware focus with their own GPU compute software stack. ROCm exists first and foremost to support AMD’s Instinct line of accelerators (used in projects such as the Frontier supercomputer), and as a result, support for non-Instinct products has been limited. Officially, AMD only supports the software stack on a pair of workstation-class RNDA 2 architecture cards (Radeon Pro W6800 & V620), while unofficial support is available for some other RDNA 2 cards and architectures – though in practice this has proven to be a mixed bag as to how reliably it works. Consequently, any announcement of new Radeon video card support for ROCm is notable, especially when it involves a consumer Radeon card.
Closing out their ROCm 5.6 announcement blog post, AMD is announcing that support for the first RDNA 3 products will be arriving in the fall. Kicking things off, the company will be adding official support for the Radeon Pro W7900 – AMD’s top workstation card – and, for the first time, the consumer-grade Radeon RX 7900 XTX. Both of these parts are based on the same RDNA 3 GPU (Navi 31), so architecturally they are identical, and it’s a welcome sign to see AMD finally embracing that and bringing a consumer Radeon card into the fold.
Broadly speaking, RDNA 3’s compute core differs significantly from RDNA 2 (and CDNA 2) thanks to the introduction of dual issue SIMD execution, and the resulting need to extract ILP from an instruction stream. So the addition of proper RDNA 3 support to the ROCm stack is not a small undertaking for AMD’s software team, especially when they are also working to support the launch of the MI300 (CDNA 3) accelerator family later this year.
Along with the first two Navi 31 cards, AMD is also committing to bringing support for “additional cards and expanded capabilities to be released over time.” To date, AMD’s official video card support has never extended beyond a single GPU within a given generation (e.g. Navi 21), so it will be interesting to see whether this means AMD is finally expanding their breadth to include more Navi 3x GPUs, or if this just means officially supporting more Navi 31 cards (e.g. W7800). AMD’s statement also seems to imply that support for the full ROCm feature set may not be available in the first iteration of RDNA 3 support, but I may be reading too much into that.
Meanwhile, though it’s not by any means official, AMD’s blog post also notes that the company is improving on their unofficial support for Radeon product, as well. Numerous issues with ROCm on unsupported GPUs have been fixed in the ROCm 5.6 release, which should make the software stack more usable on a day-to-day basis on a wider range of hardware.
Overall, this is a welcome development to see that AMD is finally lining up support for their latest desktop GPU architecture within their compute stack, as Navi 3x’s potential as a compute product has remained less than fully tapped since it launched over half a year ago. AMD has taken some not-undeserved flak over the years for ROCm’s limited support for their bread-and-butter GPU products, so this announcement, along with CEO Dr. Lisa Su’s comments earlier this month that AMD is working to improve their ROCm support, indicate that AMD is finally making some much needed (and greatly awaited) progress with improving the ROCm product stack.
Though as AMD prepares to add further hardware support for ROCm, they are also preparing to take some away, as well. Support for products based on AMD’s Vega 20 GPU, such as the Instinct MI50 and Radeon Pro VII, is set to begin sunsetting later this year. ROCm support for those products will be entering maintenance mode in Q3, with the release of ROCm 5.7, at which time no further features or performance optimizations will be added for that branch of hardware. Bug fixes and security updates will still be released for roughly another year. Ultimately, AMD is giving a formal heads up that they’re looking to drop support for that hardware entirely after Q2 of 2024.
Finally, for anyone who was hoping to see full Windows support for ROCm, despite some premature rumors, that has not happened with ROCm 5.6. Currently, AMD has a very limited degree of Windows support in the ROCm toolchain (ROCm is used for the AMD backin in both Linux and Windows editions of Blender) and ROCm development logs indicate that they’re continuing to work on the matter; but full Windows support remains elusive for the software stack. AMD has remained quite mum on the matter overall, with the company avoiding doing anything that would set any expectations for a ROCm-on-Windows release. That said, I do still expect to see proper Windows support at some point in the future, but there's nothing to indicate it's happening any time soon. Especially with MI300 on the horizon, AMD would seem to have bigger fish to fry.
In a rather unexpected move, Intel this week discontinued its Arc A770 Limited Edition graphics card, which was its flagship discrete graphics offering for desktops. Intel's partners will continue to offer their Arc A770 add-in-boards (AIBs) with 8 GB and 16 GB of GDDR6 memory.
Intel discontinues its Arc A770 Limited Edition graphics card rather abruptly: the company listed June 20, 2023 as the last product order date and the last product shipment date, which essentially means that it no longer produces and ships these boards. As soon as the remaining stock of these products will be depleted in the channel, they will no longer be available. Apparently, the product was a Limited Edition indeed since it is being EOLed less than a year on the market.
A quick check at Amazon and Newegg reveals that the Arc A770 LE board is available for as much as $497.35 at Amazon and is no longer available at Newegg. Meanwhile, Newegg has three CPU + A770 graphics cards bundles containing Acer's Predator BiFrost Arc A770 16 GB AIBs. Separately, this board costs $339 at Newegg.
Intel's own Arc A750 Limited Edition and Arc A770 Limited Edition graphics cards were meant to bring the audience the best experience possible with an all-new GPU family. In addition, they demonstrated that Intel wanted its Arc A700-series products to compete for the mainstream market segment without using fancy and huge cooling systems. Indeed, Intel's own Arc A750 LE and Arc A770 LE boards look very modest, yet provided everything that the company's ACM-G10 GPU had to offer in terms of performance and functionality.
By now, there are Intel Arc A770 graphics cards from numerous AIB producers, including ASRock, Acer, Gigabyte, Gunnir, MSI, and Sparkle, so Intel does not really need to offer its own cards to ensure that its top-of-the-line product is present on the market.
Interestingly, but Intel's Arc A750 Limited Edition graphics cards remains afloat for now. Perhaps, while Intel still has these boards in its own stock.
Source: Intel
Alongside their EPYC server CPU updates, as part of today’s AMD Data Center event, the company is also offering an update on the status of their nearly-finished AMD Instinct MI300 accelerator family. The company’s next-generation HPC-class processors, which use both Zen 4 CPU cores and CDNA 3 GPU cores on a single package, have now become a multi-SKU family of XPUs.
Joining the previously announced 128GB MI300 APU, which is now being called the MI300A, AMD is also producing a pure GPU part using the same design. This chip, dubbed the MI300X, uses just CDNA 3 GPU tiles rather than a mix of CPU and GPU tiles in the MI300A, making it a pure, high-performance GPU that gets paired with 192GB of HBM3 memory. Aimed squarely at the large language model market, the MI300X is designed for customers who need all the memory capacity they can get to run the largest of models.
First announced back in June of last year, and detailed in greater depth back at CES 2023, the AMD Instinct MI300 is AMD’s big play into the AI and HPC market. The unique, server-grade APU packs both Zen 4 CPU cores and CDNA 3 GPU cores on to a single, chiplet-based chip. None of AMD’s competitors have (or will have) a combined CPU+GPU product like the MI300 series this year, so it gives AMD an interesting solution with a truly united memory architecture, and plenty of bandwidth between the CPU and GPU tiles.
MI300 also includes on-chip memory via HBM3, using 8 stacks of the stuff. At the time of the CES reveal, the highest capacity HBM3 stacks were 16GB, yielding a chip design with a maximum local memory pool of 128GB. However, thanks to the recent introduction of 24GB HBM3 stacks, AMD is now going to be able to offer a version of the MI300 with 50% more memory – or 192GB. Which, along with the additional GPU chiplets found on the MI300X, are intended to make it a powerhouse for processing the largest and most complex of LLMs.
Under the hood, MI300X is actually a slightly simpler chip than MI300A. AMD has replaced MI300A's trio of CPU chiplets with just two CDNA 3 GPU chiplets, resulting in a 12 chiplet design overall - 8 GPU chiplets and what appears to be another 4 IO memory chiplets. Otherwise, despite excising the CPU cores (and de-APUing the APU), the GPU-only MI300X looks a lot like the MI300A. And clearly, AMD is aiming to take advantage of the synergy in offering both an APU and a flagship CPU in the same package.
Raw GPU performance aside (we don't have any hard numbers to speak of right now), a bit part of AMD's story with the MI300X is going to be memory capacity. Just offering a 192GB chip on its own is a big deal, given that memory capacity is the constraining factor for the current generation of large language models (LLMs) for AI. As we’ve seen with recent developments from NVIDIA and others, AI customers are snapping up GPUs and other accelerators as quickly as they can get them, all the while demanding more memory to run even larger models. So being able to offer a massive, 192GB GPU that uses 8 channels of HBM3 memory is going to be a sizable advantage for AMD in the current market – at least, once MI300X starts shipping.
The MI300 family remains on track to ship at some point later this year. According to AMD, the 128GB MI300A APU is already sampling to customers now. Meanwhile the 192GB MI300X GPU will be sampling to customers in Q3 of this year.
It also goes without saying that, with this announcement, AMD has solidified that they're doing a flexible XPU design at least 3 years before rival Intel. Whereas Intel scrapped their combined CPU+GPU Falcon Shores product for a pure GPU Falcon Shores, AMD is now slated to offer a flexible CPU+GPU/GPU-only product as soon as the end of this year. In this timeframe, it will be going up against products such as NVIDIA's Grace Hopper superchip, which although isn't an APU/XPU either, comes very close by linking up NVIDIA's Grace CPU with a Hopper GPU via a high bandwidth NVLink. So while we're waiting on further details on MI300X, it should make for a very interesting battle between the two GPU titans.
Overall, the pressure on AMD with regards to the MI300 family is significant. Demand for AI accelerators has been through the roof for much of the past year, and MI300 will be AMD’s first opportunity to make a significant play for the market. MI300 will not quite be a make-or-break product for the company, but besides getting the technical advantage of being the first to ship a single-chip server APU (and the bragging rights that come with it), it will also give them a fresh product to sell into a market that is buying up all the hardware it can get. In short, MI300 is expected to be AMD’s license to print money (ala NVIDIA’s H100), or so AMD’s eager investors hope.
Alongside today’s 192GB MI300X news, AMD is also briefly announcing what they are calling the AMD Infinity Architecture Platform. This is an 8-way MI300X design, allowing for up to 8 of AMD’s top-end GPUs to be interlinked together to work on larger workloads.
As we’ve seen with NVIDIA’s 8-way HGX boards and Intel’s own x8 UBB for Ponte Vecchio, an 8-way processor configuration is currently the sweet spot for high-end servers. This is both for physical design reasons – room to place the chips and room to route cooling through them – as well as the best topologies that are available to link up a large number of chips without putting too many hops between them. If AMD is to go toe-to-toe with NVIDIA and to capture part of the HPC GPU market, then this is one more area where they’re going to need to match NVIDIA’s hardware offerings
AMD is calling the Infinity Architecture Platform an “industry-standard” design. Accoding to AMD, they're using an OCP server platform as their base here; and while this implies that MI300X is using an OAM form factor, we're still waiting to get explicit confirmation of this.
Intel this week introduced its highest performing products aimed at professional visualization market: the Arc Pro A60 12 GB graphics card for desktops as well as the Arc Pro A60M graphics processor for notebooks. Both workstations-aimed units use the company's ACM-G10 GPU but neither use it fully enabled version with maximum performance.
Intel's Arc Pro A60 for desktops and Arc Pro A60M for laptops rely on the company's biggest discrete GPU — the ACM-G10 — with 16 Xe cores (2048 stream processors), 16 ray tracing units, and 256 vector engines meaning that both graphics solutions enable only half the resources the graphics processor has. As far as memory subsystems are concerned, the desktop graphics board comes with 12 GB of GDDR6 memory and features a 192-bit interface, whereas the mobile solution is equipped with 8 GB of GDDR6 memory connected to the GPU using a 128-bit bus. The desktop card is rated for up to 130W TGP and uses a single-slot cooling system.
Given that both Arc Pro A60 and Arc Pro A60M rely on severely cut-down ACM-G10 GPU, their compute performance is at least two times lower when compared the company's range-topping Arc A770 and Arc A770M products. Yet, these are the highest-performing workstation-grade graphics offerings from Intel to date as other Arc Pro-series devices use the much less capable ACM-G11 GPU with fewer Xe cores and ray tracing engines.
Intel positions these graphics solutions for a range of applications, including architecture, computer aided design, engineering, and manufacturing industries just to name a few. As such, these boards will come with quarterly ISV-certified driver releases for regular performance optimizations in programs developed by Autodesk, Bentley, Dassault Systèmes, Nemetschek, PTC, and Siemens. This is a notable development as,
Also, all Arc Pro graphics solutions feature full-fat media engines supporting hardware encoding/decoding of AV1, H.264, H.265, and VP9 streams, whereas desktop boards all come with four DisplayPort 2.0 outputs with UHBR10 that support resolutions of up to 8Kp60 with DSC.
The Intel Arc Pro A60 graphics card will soon be available through official Intel distributors, the company said Meanwhile, the mobile variant, the Intel Arc Pro A60M, is set to find its place in laptops from popular workstation OEMs within the next few months, the company indicated.
Without much fanfare, Biostar has introduced its first graphics card based on Intel Arc graphics processors. The add-in-board is aimed at entry-level gaming PCs, and is admittedly not very remarkable itself. But the fact that Intel has a new AIB partner, and that Biostar now has graphics cards powered by GPUs from all three major vendors, are important developments for the wider industry.
The Biostar Intel Arc A380 graphics card demonstrated at Computex 2023 is based on the ACM-G11 processor and features a minimalistic design with a rather simplistic single-fan cooler. The AIB does not need any auxiliary PCIe power connectors and will fit into the vast majority of desktops that are new or already in use (including Mini-ITX one), so it can be used both for new PCs and for upgrades. It is unclear whether Biostar plans to offer these products worldwide.
Biostar 🤜🤛 Intel Arc#Computex23 pic.twitter.com/wpJ1jBg4Zs
— Hassan Mujtaba (@hms1193) May 31, 2023
The board first appeared in a Biostar video posted on May 31, 2023, and then was spotted at the company's Computex booth by a WCCFTech editor Hassan Mujtaba.
While Intel is one of the world's most recognized brands, its modern entry into the discrete GPU game has not been met with a lot of enthusiasm from the top video card manufacturers. As of today, Intel has a number of partners building AIBs powered by its standalone graphics processors, including ASRock, Acer, Gigabyte, Gunnir, MSI, and Sparkle. The addition of Biostar in this case seems like an important event for Intel's GPU business, especially in going after the entry-level segment of the market.
Speaking of Biostar, it is noteworthy that while the company has introduced its Radeon RX 7900-series and Radeon RX 7600-series graphics cards, it yet has to offer any new products based on NVIDIA's GeForce RTX 40-series GPUs. In fact, it is unclear whether the company has plans for any of these at all and its presence at Computex has not shed any light on its intentions.
Asus has introduced a new flagship RTX 4090 graphics card that uses an all-in-one liquid cooling system combined with liquid metal thermal interface. Dubbed the ROG Matrix GeForce RTX 4090, Asus says that its advanced cooler combined with extremely efficient thermal interface will ensure the maximum boost clocks possible, with Asus taking clear aim of producing the fastest gaming graphics card on the market.
Proper power delivery and efficient cooling are main ways to enable consistently high CPU and GPU performance these days, so when designing its ROG Matrix GeForce RTX 4090, the company used its own proprietary printed circuit board (PCB) with an advanced voltage regulating module (VRM). Meanwhile cooling is being provided by an all-in-one liquid cooling system that removes heat not only from GPU, but also from memory and VRM, exhausting that heat via the attached "extra-thick" 360mm radiator.
But Asus says that its ROG Matrix GeForce RTX 4090 has a secret ingredient that its rivals lack: liquid metal thermal interface material (TIM) that ensures superior heat transfer from hot components to cooling systems.
Asus does not disclose what type of liquid metal TIM it uses for graphics cards (it uses ThermalGrizzly's Conductonaut Extreme for some laptops), bus usually such thermal interfaces are made from gallium or gallium alloys, which are liquid at or near room temperature and are great conductors of heat.
But there are also some risks and challenges associated with using liquid metal thermal interfaces. Firstly, they are electrically conductive, which means that if the material spills or is not properly contained, it could cause a short circuit. Secondly, these materials can be corrosive to certain metals like aluminum. Thirdly, applying liquid metal can be more complicated than using other types of thermal paste, requiring careful handling and precision.
Asus says that it has been using liquid metal TIMs in its laptops for years, so using them for graphics cards does not seem to be a big challenge for the company.
Image Credit: Future/TechRadar
Asus is not disclosing the complete specifications of the ROG Matrix GeForce RTX 4090 for the moment, but it certainly hopes to make the graphics card the world's fastest. It remains to be seen whether the product will indeed be the fastest out-of-box, but it will certainly offer a noteworthy overclocking potential when compared to regular GeForce RTX 4090 graphics boards with regular coolers.
The Asus ROG Matrix GeForce RTX 4090 will be a limited-edition card available for sale in Q3.
Teeing off an AI-heavy slate of announcements for NVIDIA, the company has confirmed that their Grace Hopper “superchip” has entered full production. The combination of a Grace CPU and Hopper H100 GPU, Grace Hopper is designed to be NVIDIA’s answer for customers who need a more tightly integrated CPU + GPU solution for their workloads – particularly for AI models.
In the works for a few years now, Grace Hopper is NVIDIA’s efforts to leverage both their existing strength in the GPU space and newfound efforts in the CPU space to deliver a semi-integrated CPU/GPU product unlike anything their top-line competitors offer. With NVIDIA’s traditional dominance in the GPU space, the company has essentially been working backwards, combining their GPU technology with other types of processors (CPUs, DPUs, etc) in order to access markets that benefit from GPU acceleration, but where fully discrete GPUs may not be the best solution.
| NVIDIA Grace Hopper Specifications | |||
| Grace Hopper (GH200) | |||
| CPU Cores | 72 | ||
| CPU Architecture | Arm Neoverse V2 | ||
| CPU Memory Capacity | <=480GB LPDDR5X (ECC) | ||
| CPU Memory Bandwidth | <=512GB/sec | ||
| GPU SMs | 132 | ||
| GPU Tensor Cores | 528 | ||
| GPU Architecture | Hopper | ||
| GPU Memory Capcity | <=96GB | ||
| GPU Memory Bandwidth | <=4TB/sec | ||
| GPU-to-CPU Interface | 900GB/sec NVLink 4 |
||
| TDP | 450W - 1000W | ||
| Manufacturing Process | TSMC 4N | ||
| Interface | Superchip | ||
In this first NVIDIA HPC CPU + GPU mash-up, the Hopper GPU is the known side of the equation. While it only started shipping in appreciable volumes this year, NVIDIA was detailing the Hopper architecture and performance expectations over a year ago. Based on the 80B transistor GH100 GPU, H100 brings just shy of 1 PFLOPS of FP16 matrix math throughput for AI workloads, as well as 80GB of HBM3 memory. H100 is itself already a huge success – thanks to the explosion of ChatGPT and other generative AI services, NVIDIA is already selling everything they can make – but NVIDIA is still pushing ahead with their efforts to break into markets where the workloads require closer CPU/GPU integration.
Being paired with H100, in turn, is NVIDIA’s Grace CPU, which itself just entered full production a couple of months ago. The Arm Neoverse V2-based chip packs 72 CPU cores, and comes with up to 480GB of LPDDR5X memory. And while the CPU cores are themselves plenty interesting, the bigger twist with Grace has been NVIDIA’s decision to co-package the CPU with LPDDR5X, rather than using slotted DIMMs. The on-package memory has allowed NVIDIA to use both higher clocked and lower power memory – at the cost of expandability – which makes Grace unlike any other HPC-class CPU on the market. And potentially a very big deal for Large Language Model (LLM) training, given the emphasis on both dataset sizes and the memory bandwidth needed to shuffle that data around.
It’s that data shuffling, in turn, that helps to define a single Grace Hopper board as something more than just a CPU and GPU glued together on the same board. Because NVIDIA equipped Grace with NVLink support – NVIDIA’s proprietary high-bandwidth chip interconnect – Grace and Hopper have a much faster interconnect than a traditional, PCIe-based CPU + GPU setup. The resulting NVLink Chip-to-Chip (C2C) link offers 900GB/second of bandwidth between the two chips (450GB/sec in each direction), giving Hopper the ability to talk back to Grace even faster than Grace can read or write to its own memory.
The resulting board, which NVIDIA calls their GH200 “superchip”, is meant to be NVIDIA’s answer to the AI and HPC markets for the next product cycle. For customers who need a more local CPU than a traditional CPU + GPU setup – or perhaps more pointedly, more quasi-local memory than a stand-alone GPU can be equipped with – Grace Hopper is NVIDIA’s most comprehensive compute product yet. Meanwhile, with there being some uncertainty over just how prevalent the Grace-only (CPU-only) superchip will be, given that NVIDIA is currently on an AI bender, Grace Hopper may very well end up being where we see the most of Grace, as well.
According to NVIDIA, systems incorporating GH200 chips are slated to be available later this year.
Meanwhile, even though Grace Hopper is not technically out the door yet, NVIDIA is already at work building its first DGX system around the chip. Though in this case, “DGX” may be a bit of a misnomer for the system, which unlike other DGX systems isn’t a single node, but rather a full-on multi-rack computational cluster – hence NVIDIA terming it a “supercomputer.”
At a high level, the DGX GH200 AI Supercomputer is a complete, turn-key, 256 node GH200 cluster. Spanning some 24 racks, a single DGX GH200 contains 256 GH200 chips – and thus, 256 Grace CPUs and 256 H100 GPUs – as well as all of the networking hardware needed to interlink the systems for operation. In cumulative total, a DGX GH200 cluster offers 120TB of CPU-attached memory, another 24TB of GPU-attached memory, and a total of 1 EFLOPS of FP8 throughput (with sparsity).
Look Closer: That's Not a Server Node - That's 24 Server Racks
Linking the nodes together is a two-layer networking system built around NVLink. 96 local, L1 switches provide immediate communications between the GH200 blades, while another 36 L2 switches provide a second layer of connectivity tying together the L1 switches. And if that’s not enough scalability for you, DGX GH200 clusters can be further scaled up in size by using InfiniBand, which is present in the cluster as part of NVIDIA’s use of ConnectX-7 network adapters.
The target market for the sizable silicon cluster is training large AI models. NVIDIA is leaning heavily on their existing hardware and toolsets in the field, combined with the sheer amount of memory and memory bandwidth a 256-node cluster affords to be able to accommodate some of the largest AI models around. The recent explosion in interest in large language models has exposed just how much memory capacity is a constraining factor, so this is NVIDIA’s attempt to offer a single-vendor, integrated solution for customers with especially large models.
And while not explicitly disclosed by NVIDIA, in a sign that they all pulling out all of the stops for the DGX GH200 cluster, the memory capacities they’ve listed indicate that NVIDIA isn’t just shipping regular H100 GPUs as part of the system, but rather they are using their limited availability 96GB models, which have the normally-disabled 6th stack of HBM3 memory enabled. So far, NVIDIA only offers these H100 variants in a handful of products – the specialty H100 NVL PCIe card and now in some GH200 configurations – so DGX GH200 is slated to get some of NVIDIA’s best silicon.
Of course, don’t expect a supercomputer from NVIDIA to come cheaply. While NVIDIA is not announcing any pricing this far in advance, based on HGX H100 board pricing (8x H100s on a carrier board for $200K), a single DGX GH200 is easily going to cost somewhere in the low 8 digits. Suffice it to say, DGX GH200 is aimed at a rather specific subset of Enterprise clientele – those who need to do a lot of large model training and have the deep pocketbooks to pay for a complete, turn-key solution.
Ultimately, however, DGX GH200 isn’t just meant to be a high-end system for NVIDIA to sell to deep-pocketed customers, but it’s the blueprint for helping their hyperscaler customers build their own GH200-based clusters. Building such a system is, after all, the best way to demonstrate how it works and how well it works, so NVIDIA is forging their own path in this regard. And while NVIDIA would no doubt be happy to sell a whole lot of these DGX systems directly, so long as it gets hyperscalers, CSPs, and others adopting GH200 in large numbers (and not, say, rival products), then that’s still going to be a win in NVIDIA’s books.
In the meantime, for the handful of businesses that can afford a DGX GH200 AI Supercomputer, according to NVIDIA the systems will be available by the end of the year.
Closing out the most recent earnings season for the PC industry is, as always, NVIDIA. The company’s unusual, nearly year-ahead fiscal calendar means that they get the benefit of being casually late in reporting their results. And in this case, they’ve ended up being the proverbial case of saving the best for last.
For the first quarter of their 2024 fiscal year, NVIDIA booked $7.2 billion in revenue, which is a 13% drop over the year-ago quarter. Like the rest of the chip industry, NVIDIA has been weathering a significant slump in demand for computing products over the past few quarters, which in turn has dented NVIDIA’s revenue and profitability. However, while NVIDIA’s consumer-focused gaming division has continued to take matters on the chin, the strong performance of NVIDIA’s data center group has kept the company as a whole fairly profitable, with the most recent quarter setting a segment record and helping NVIDIA to avoid the tough financial situations faced by rivals AMD and Intel.
| NVIDIA Q1 FY2024 Financial Results (GAAP) | |||||
| Q1 FY2024 | Q4 FY2023 | Q1 FY2023 | Q/Q | Y/Y | |
| Revenue | $7.2B | $6.1B | $8.3B | +19% | -13% |
| Gross Margin | 64.6% | 63.3% | 65.5% | +1.3ppt | -0.9ppt |
| Operating Income | $2.1B | $1.3B | $1.9B | +70% | +15% |
| Net Income | $2.0B | $1.4B | $1.6B | +44% | +26% |
| EPS | $0.82 | $0.57 | $0.64 | +44% | +28% |
To that end, while Q1’FY24 was not by any means a record quarter for NVIDIA, it was still a relatively strong one for the company. NVIDIA’s net income of $2 billion makes for one of their better quarters in that regard, and it’s actually up 26% year-over-year despite the revenue drop. That said, reading between the lines will find that NVIDIA paid their Arm acquisition breakup fee last year (Q1’FY23), so NVIDIA’s GAAP net income looks a bit better than it otherwise would; while non-GAAP net income would be down 21%. Meanwhile, NVIDIA’s gross margins have held strong in the most recent quarter, with NVIDIA posting a GAAP gross margin of 64.6%.
But even a solid quarter during an industry slump is arguably not the biggest news to come out of NVIDIA’s most recent earnings report. Rather, it’s the company’s projections for Q2’FY24. In short, NVIDIA is expecting revenue to explode in Q2, with the company forecasting $11 billion in sales. Should it come to fruition, such a quarter would blow well past NVIDIA’s previous revenue records – and shattering Wall Street expectations. As a result, NVIDIA’s stock has already taken off in overnight trading, and by the time the market opens a bit later this morning, NVIDIA is expected to be a $930B+ company, knocking on the door of crossing a market capitalization of a trillion dollars.
Asus has expanded the company's GeForce RTX 40-series product portfolio with two new RTX 4090 graphics cards. The ROG Strix LC GeForce RTX 4090 and TUF Gaming GeForce RTX 4090 OC, which are available in regular and OC editions, have arrived to compete in the high-end segment. What makes these cards notable, in turn, is their reduced size: the new cards are physically smaller than Asus' early RTX 4090 offerings, as well as many of the competitors on the market.
The GeForce RTX 4090 is a 450W gaming graphics card, with large coolers to match. Even NVIDIA's hard-to-get GeForce RTX 4090 Founders Edition is a triple-slot graphics card, and air-cooled AIB cards tend to be larger still. So for the size-conscious gamer, this leaves liquid cooled cards, which brings us to Asus's new ROG Strix LC GeForce RTX 4090. The closed-loop card moves a lot of its bulk off to an attached 240 mm radiator block, bringing the card itself down to 2.6-slots wide.
The ROG Strix LC GeForce RTX 4090's hybrid cooling system packs a cold plate that cools the large AD102 GPU and neighboring GDDR6X memory chips. The heat is transferred to the 240 mm radiator through 560 mm tubing, so there won't be an issue with large cases. A low-profile heatsink with a blower-style cooling fan keeps the other power delivery components cool. Meanwhile the radiator itself is equipped with a pair of 120 mm ARGB cooling fans are present to dissipate the heat once it gets there.
| Asus GeForce RTX 4090 Specifications | |||
| AnandTech | ROG Strix LC GeForce RTX 4090 | TUF Gaming GeForce RTX 4090 OG | TUF Gaming GeForce RTX 4090 |
| Regular Edition Boost Clock (Default / OC) |
2,520 MHz / 2,550 MHz | 2,520 MHz / 2,550 MHz | 2,520 MHz / 2,550 MHz |
| OC Edition Boost Clock (Default / OC) |
2,610 MHz / 2,640 MHz | 2,565 MHz / 2,595 MHz | 2,565 MHz / 2,595 MHz |
| Display Outputs | 2 x HDMI 2.1a 3 x DisplayPort 1.4a |
2 x HDMI 2.1a 3 x DisplayPort 1.4a |
2 x HDMI 2.1a 3 x DisplayPort 1.4a |
| Design | 2.6 Slot | 3.2 Slot | 3.65 Slot |
| Power Connectors | 1 x 16-pin | 1 x 16-pin | 1 x 16-pin |
| Dimensions | 293 x 133 x 52 mm |
325.9 x 140.2 x 62.8 mm | 348.2 x 150 x 72.6 mm |
| Radiator Dimensions | 272 x 121 x 54 mm | N/A | N/A |
Asus's other new RTX 4090 card, the air-cooled TUF Gaming GeForce RTX 4090 OG, is a unique case of its own. Technically, it's a new SKU; however, the graphics card reuses the TUF Gaming cooler from the TUF Gaming GeForce RTX 3090 Ti.
This is notable because the TUF cooler used on the 3090 Ti was a good bit smaller than Asus's first RTX 4090 cooler. The net result is that these changes bring the new OG card's width from 3.65-slots (and arguably, wide enough that you need to leave a 4th slot open for air flow) down to 3.2 slots - just enough room for proper airflow if the neighboring 4th sot is occupied. Altogether, the OG model is smaller in every dimension, shaving off 6% of its height, 7% of its length, and 13% of its width. Asus doesn't list the weight of its graphics cards, so we cannot comment on whether the new OG version has lost weight.
By most accounts, Asus's current RTX 4090 cooler is highly effective – it's just also really big. So offering a separate SKU with a smaller cooler makes a good deal of sense, especially given how popular NVIDIA's true triple-slot Founders Edition card has been. The smaller TUF cooler is rated for the same 450W TDP as the larger TUF 4090 cooler, but, as always, there may be performance/acoustic tradeoffs involved.
There's one other change that Asus doesn't advertise with the TUF Gaming GeForce RTX 4090 OG. The renders on the product page show the graphics card with a longer PCB. One of the advantages of the more compact PCB on the previous model was that it permitted Asus (and NVIDIA) to vent heat out of the back side of the card, as well as to optimize the trace layouts and component placement. Meanwhile, with the longer PCB, Asus relocated the 16-pin power connector. Instead of being placed in the middle, the power connector is on the farther right side.
Between the two new cards, the ROG Strix LC GeForce RTX 4090 ends up with the edge in clockspeeds, flaunting boost clock speeds up to 2,640 MHz when in its highest performance mode. Meanwhile, the TUF Gaming GeForce RTX 4090 OG series have the same clock speeds as the vanilla models, with a rated boost clock of 2520 MHz stock and 2595 MHz when the OC card is in its highest mode. In addition, the ROG Strix LC GeForce RTX 4090 and TUF Gaming GeForce RTX 4090 OG have other attributes in common, including using a single 16-pin power connector and a display output layout consisting of two HDMI 2.1a ports and three DisplayPort 1.4a outputs.
Asus hasn't revealed the pricing or availability of the new graphics cards. For reference, the TUF Gaming GeForce RTX 4090 and OC Edition retail for $1,599 and $1,799, respectively. The OG counterparts likely have similar price tags. Meanwhile, we'd expect the ROG Strix LC GeForce RTX 4090 to carry a more considerable premium due to the AIO liquid cooling design.
NVIDIA is launching a new game bundle for its latest generaiton GeForce RTX 40-series graphics cards and OEM systems. This time, NVIDIA has teamed up with Activision Blizzard to offer a free copy of the latest iteration of their wildly popular action RPG series, Diablo IV.
This promotion will run globally, starting now and running through June 16, 2023. For more than a month, customers purchasing GeForce RTX 4090, 4080, 4070 Ti, 4070 graphics cards or desktops containing one of them from various vendors will get a free digital download code of Diablo IV Standard Edition on Battle.net. The code for the title must be redeemed before July 13, 2023.
| NVIDIA Current Game Bundles (May 2023) |
|
| Video Card (incl. systems and OEMs) |
Game |
| GeForce RTX 40 Series Desktop (All) | Diablo IV |
| GeForce RTX 30 Series Desktop (All) | None |
| GeForce RTX 40 Series Laptop (All) | None |
| GeForce RTX 30 Series Laptop (All) | None |
For NVIDIA, Diablo IV will also be a technology showcase, as it is set to support the DLSS 3 upscaling technology as well as the Reflex latency cutting out-of-box at launch. Ray tracing is also slated to be added at some point after the game launches. At retail pricing, Activision Blizzard's Diablo IV Standard Edition costs $69.99 at Battle.net, though NVIDIA is undoubtedly getting a bulk deal.
It should be noted that this latest game bundle is just for NVIDIA's RTX 40 series desktop cards. Unlike the since-expired Redfall bundle, NVIDIA is not offering Diablo IV (or any other games) with GeForce-based laptops. Nor are any remaining GeForce RTX 30 series producted covered.
Diablo IV will officially release on June 4, 2023.
Source: NVIDIA
In a brief note posted to its investor relations portal this morning, AMD has announced that they will be holding a special AI and data center-centric event on June 13th. Dubbed the “AMD Data Center and AI Technology Premiere”, the live event is slated to be hosted by CEO Dr. Lisa Su, and will be focusing on AMD’s AI and data center product portfolios – with a particular spotlight on AMD’s expanded product portfolio and plans for growing out these market segments.
The very brief announcement doesn’t offer any further details on what content to expect. However, the very nature of the event points a clear arrow at AMD’s forthcoming Instinct Mi300 accelerator. MI300 is AMD’s first shot at building a true data center/HPC-class APU, combining the best of AMD’s CPU and GPU technologies. AMD has offered only a handful of technical details about MI300 thus far – we know it’s a disaggregated design, using multiple chiplets built on TSMC’s 5nm process, and using 3D die stacking to place them over a base die – and with MI300 slated to ship this year, AMD will need to fill in the blanks as the product gets closer to launch.
As we noted in last week’s AMD earnings report, AMD’s major investors have been waiting with baited breath for additional details on the accelerator. Simply put, investors are treating data center AI accelerators as the next major growth opportunity for high-performance silicon – eyeing the high margins these products have afforded over at NVIDIA and other AI-adjacent rivals – so there is a lot of pressure on AMD to claim a slice of what’s expected to be a highly profitable pie. MI300 is a product that has been in the works for years, so the pressure is more of a reaction to the money than the silicon itself, but still, MI300 is expected to be AMD’s best opportunity yet to capture a meaningful portion of the data center GPU market.
MI300 aside, given the dual AI and data center focus of the event, this is also where we’re likely to see more details on AMD’s forthcoming EPYC “Genoa-X” CPUs. The L3 V-Cache-equipped version of AMD’s current-generation EPYC 9004 series Genoa CPUs, Genoa-X has been on AMD’s roadmap for a while. And with their consumer equivalent parts already shipping (Ryzen 7000X3D), AMD should be nearing completion of the EPYC parts. AMD has previously confirmed that Genoa-X will ship with up to 96 CPU cores, with over 1GB in total L3 cache available on the chip to further boost performance on workloads that benefit from the extra cache.
AMD’s ultra-dense EPYC Bergamo chip is also in the pipeline, though given the high-performance aspects of the presentation, it’s a little more questionable whether it will be at the show. Based on AMD’s compacted Zen4c architecture, Bergamo is aimed at cloud service providers who need lots of cores to split up amongst customers, with up to 128 CPU cores on a single Bergamo chip. Like Genoa-X, Bergamo is slated to launch this year, so further details about it should come to light sooner than later.
But whatever AMD does (or doesn’t) show at their event, we’ll find out on June 13th at 10am PT (17:00 UTC). AMD will be live streaming the event from their website as well as YouTube.
Lateo.net - Flux RSS en pagaille (pour en ajouter : @ moi) 