After posting a teaser video last week, MSI has followed up by announcing their latest ultra-high-end Lightning-branded graphics card: the MSI GeForce GTX 1080 Ti LIGHTNING Z. The triple-slot-width, triple-fan, and triple-8-pin power connector card co…
Similar to last year, at this year’s International Supercomputing Conference (ISC) NVIDIA has announced and detailed a PCI Express version of their latest Tesla GPU accelerator, the Volta-based V100. The conference itself runs from June 19 to 22, and with several speakers from NVIDIA scheduled for events tomorrow, NVIDIA is set to outline its next-generation efforts in HPC and deep learning with Volta.
With Volta discussed and described at their GPU Technology Conference in mid-May, NVIDIA upped the ante in terms of both features and reticle size: V100 is 815mm2 of custom TSMC 12FFN silicon, chock full of tensor cores and unified L1 cache per SM, along with many more fundamental – and as of yet not fully revealed – microarchitectural changes.
Like the previous Pascal iteration, the Tesla V100 PCIe offers a more traditional form factor as opposed to NVIDIA’s own mezzanine-type SXM2 form factor. This allows vendors to drop Tesla cards in traditional PCIe systems, making the cards far more accessible to server builders who don’t want to build around NVIDIA’s SXM2 connector or carrier board. The tradeoff being that the PCIe cards have a lower 250W TDP, and they don’t get NVLink, instead relying on just PCIe.
| NVIDIA Tesla Family Specification Comparison | ||||||
| Tesla V100 (SXM2) |
Tesla V100 (PCIe) |
Tesla P100 (SXM2) |
Tesla P100 (PCIe) |
|||
| CUDA Cores | 5120 | 5120 | 3584 | 3584 | ||
| Tensor Cores | 640 | 640 | N/A | N/A | ||
| Core Clock | ? | ? | 1328MHz | ? | ||
| Boost Clock(s) | 1455MHz | ~1370MHz | 1480MHz | 1300MHz | ||
| Memory Clock | 1.75Gbps HBM2 | 1.75Gbps HBM2 | 1.4Gbps HBM2 | 1.4Gbps HBM2 | ||
| Memory Bus Width | 4096-bit | 4096-bit | 4096-bit | 4096-bit | ||
| Memory Bandwidth | 900GB/sec | 900GB/sec | 720GB/sec | 720GB/sec | ||
| VRAM | 16GB | 16GB | 16GB | 16GB | ||
| L2 Cache | 6MB | 6MB | 4MB | 4MB | ||
| Half Precision | 30 TFLOPS | 28 TFLOPS | 21.2 TFLOPS | 18.7 TFLOPS | ||
| Single Precision | 15 TFLOPS | 14 TFLOPS | 10.6 TFLOPS | 9.3 TFLOPS | ||
| Double Precision | 7.5 TFLOPS (1/2 rate) |
7 TFLOPS (1/2 rate) |
5.3 TFLOPS (1/2 rate) |
4.7 TFLOPS (1/32 rate) |
||
| Tensor Performance (Deep Learning) |
120 TFLOPS | 112 TFLOPS | N/A | N/A | ||
| GPU | GV100 (815mm2) | GV100 (815mm2) | GP100 (610mm2) | GP100 (610mm2) | ||
| Transistor Count | 21B | 21B | 15.3B | 15.3B | ||
| TDP | 300W | 250W | 300W | 250W | ||
| Form Factor | Mezzanine (SXM2) | PCIe | Mezzanine (SXM2) | PCIe | ||
| Cooling | Passive | Passive | Passive | Passive | ||
| Manufacturing Process | TSMC 12nm FFN | TSMC 12nm FFN | TSMC 16nm FinFET | TSMC 16nm FinFET | ||
| Architecture | Volta | Volta | Pascal | Pascal | ||
On the surface, the addition of tensor cores is the most noticeable change. To recap, tensor cores can be liked to a series of unified ALUs that are able to multiply two 4×4 FP16 matrices together and subsequently add that product to an FP16 or FP32 4×4 matrix in a fused multiply add operation, as opposed to conventional FP32 or FP64 CUDA cores. In the end, this means that for very specific kinds (and specifically programmed) workloads, Volta can take advantage of the 100+ TFLOPS capability that NVIDIA has tossed into the mix.
As for the specific specifications of the PCIe Tesla V100, it’s similarly configured to the SXM2 version, getting the same number of CUDA cores and memory capacity, however operating at a lower clockspeed in-line with its reduced 250W TDP. Based on NVIDIA’s throughput figures, this puts the PCIe card’s boost clock at around 1370MHz, 85MHz (~6%) slower than the SXM2 version.
Interestingly, unlike the Tesla P100 family, NVIDIA isn’t offering a second-tier PCIe card based on salvaged chips; so this generation doesn’t have an equivalent to the 12GB PCIe Tesla P100. NVIDIA’s experience with GP100/interposer/HBM2 assembly as well as continuing production of HBM2 has likely reduced the need for memory-salvaged parts.
Finally, PCIe-based Tesla V100 accelerators are “expected to be available later this year from NVIDIA reseller partner and manufacturers,” including Hewlett Packard Enterprise, which will offer three different PCIe Volta systems.
Eagerly anticipated for later this month is the launch of AMD’s first wave of Radeon Vega cards, the first-run workstation/early adopter-focused Radeon Vega Frontier Edition. To date, AMD has not yet said anything further about the launch since last month’s Computex unveil, however it appears that either AMD is opting to quietly release the sure to sell out cards, or some of their retailers have jumped the gun, as listings for both models have begun to show up.
SabrePC, one of the industry’s more specialized retailers whom tends to focus on workstation and server products, has posted listings for both of the Radeon Vega Frontier Edition cards that AMD has previously unveiled. That is, both the air-cooled card and the closed loop liquid cooled model. As you’d expect for these early-run cards, they won’t come cheap: the air cooled model is listed at $1199, while the liquid cooled card is higher still at $1799.
As a matter of editorial policy I don’t typically post news about retailer listings; these are often erroneous, or at the very least speculative. However any listings at SabrePC raise an eyebrow as they’re a more straight-laced player and one of the traditional retailers for workstation products. So they’re not known to post faulty listings. Which, coupled with the fact that other workstation retailers are also listing these cards, leads me to believe that this week’s listing isn’t an accident, even if AMD themselves aren’t saying more about the product.
In any case, we had no real guidance for where AMD would price these cards at prior to today, so I’m admittedly a bit surprised to see the Frontier Edition cards come in as (relatively) cheap as they have. $1199 for the air cooled card is less than similar NVIDIA Quadros (and Radeon Pro cards, for that matter), and is perfectly aligned with NVIDIA Titan Xp pricing. Meanwhile the liquid cooled card is a bit more surprising with its $600 premium. All messaging so far from AMD is that these are a low volume part meant for customers to evaluate Vega as early as possible, so it’ll be interesting to see where AMD goes from here.
| AMD Workstation Card Specification Comparison | ||||||
| Radeon Vega Frontier Edition (Unconfirmed) |
Radeon Pro Duo (Polaris) | Radeon Pro WX 7100 | Radeon Fury X | |||
| Stream Processors | 4096 | 2 x 2304 | 2304 | 4096 | ||
| Texture Units | ? | 2 x 144 | 144 | 256 | ||
| ROPs | 64? | 2 x 32 | 32 | 64 | ||
| Boost Clock | 1.6GHz | 1243MHz | 1243MHz | 1050MHz | ||
| Single Precision | 13.1 TFLOPS | 11.5 TFLOPS | 5.7 TFLOPS | 8.6 TFLOPS | ||
| Half Precision | 26.2 TFLOPS | 11.5 TFLOPS | 5.7 TFLOPS | 8.6 TFLOPS | ||
| Memory Clock | 1.89Gbps HBM2 | 7Gbps GDDR5 | 7Gbps GDDR5 | 1Gbps HBM | ||
| Memory Bus Width | 2048-bit | 2 x 256-bit | 256-bit | 4096-bit | ||
| Memory Bandwidth | 483GB/sec | 2x 224GB/sec | 224GB/sec | 512GB/sec | ||
| VRAM | 16GB | 2 x 16GB | 8GB | 4GB | ||
| Typical Board Power | ? | 250W | 130W | 275W | ||
| GPU | Vega (1) | Polaris 10 | Polaris 10 | Fiji | ||
| Architecture | Vega | Polaris | Polaris | GCN 1.2 | ||
| Manufacturing Process | GloFo 14nm | GloFo 14nm | GloFo 14nm | TSMC 28nm | ||
| Launch Date | 06/2017 | 05/2017 | 10/2016 | 06/24/15 | ||
| Launch Price | Air: $1199 Liquid: $1799 |
$999 | $649 | $649 | ||
Meanwhile SabrePC also lists technical specifications for the Frontier Edition cards, with both cards listed at the same memory bandwidth and peak throughput. At 13.1 TFLOPS FP32, this would put the GPU clockspeed at 1.6GHz on the dot, just a smidge higher than AMD’s own presentations last month. Meanwhile 483GB/sec of memory bandwidth puts the memory clock at just under 1.9Gbps. That both cards are listed with the same specifications is a bit surprising, and given the price difference I’m not wholly convinced that Sabre has the right specifications for the cheaper air cooled card – distinctly cheaper cards are usually built around harvested processors – but for now it’s what we have to work with. It may very well be that the listings are correct, but the air cooled card is expected to throttle more often relative to the high-efficiency air cooler.
In the meantime I’ve reached out to AMD for more information on these new listings, particularly since AMD’s official Frontier Edition release isn’t slated to be until the 27th. However quiet nature of these listings does have me wondering if AMD is purposely looking to avoid additional press at the moment – opting to silently get them into the hands of distributors to get out to their professional customers – as the company had made it clear that they’re not aiming these cards at consumers.
Even though the major US national laboratories are just now starting to take delivery of the supercomputers they ordered a few years back, due to the long and complex development process for these projects, the US Department of Energy(DOE) has already…
NVIDIA has announced yet another game bundle, this time pairing Bungie’s much-anticipated Destiny 2 with NVIDIA’s highly performant GeForce GTX 1080 and 1080 Ti. Systems (including NVIDIA’s own GeForce GTX Battlebox) and laptops cont…