Jensen Huang, CEO of Nvidia, has announced the company will be making a complete transition to Intel processors(opens in new tab) for its upcoming DGX H100 unit and supercomputer projects in the future. Nvidia will be using Intel’s forthcoming Rapids Sapphire Xeon processor lineup as a total replacement for AMD’s Zen 3 EPYC CPU, which Nvidia has been using extensively for years.
Huang says the primary reason for switching CPU brands was the exceptional single-threaded performance Sapphire Rapids offers over the competition. …
Sapphire Rapids is Intel’s next-generation server architecture and will use the same Golden Cove performance cores as Intel’s desktop Alder Lake architecture. Sapphire Rapids is the server version of Alder Lake.
As a result, Sapphire Rapids should have a similar IPC uplift as Intel’s Alder Lake architecture. Sapphire Rapids will also support the latest memory and storage technologies, including DDR5, HBM2E, and PCIe Gen 5, bringing it up to date with the latest and most remarkable technologies.
…
Sapphire Rapids will also support several other newer technologies Intel’s previous Ice Lake server architecture did not. Some of these include the new AMX AVX512_BF16 instruction set for deep learning applications and a special Data Streaming Accelerator that offloads all storage calls from the CPU to a dedicated chip to decrease CPU utilization.
For Nvidia, Sapphire Rapids will provide its DGX units with far higher single-threaded performance plus higher memory bandwidth and more PCIe bandwidth compared to AMD’s current Zen 3 offerings.
DGX H100 caters to AI-intensive applications in particular, with each DGX unit featuring 8 of Nvidia’s brand new Hopper H100 GPUs with a performance output of 32 petaFlops. For more details, check out our coverage.
From your quote: Jensen Huang, CEO of Nvidia, has announced the company will be making a complete transition to Intel processors(opens in new tab) for its upcoming DGX H100 unit and supercomputer projects in the future. Nvidia will be using Intel’s forthcoming Rapids Sapphire Xeon processor lineup as a total replacement for AMD’s Zen 3 EPYC CPU, which Nvidia has been using extensively for years.
So, is Intel (NASDAQ: INTC) actually producing a product that’s better suited to computer graphics, or is Advanced Micro Devices (NYSE: AMD) not able to deliver the quantity of processors that Nvidea needs, or has Nvidia has jumped off of a cliff?
Methinks that AMD makes more profit selling processors for supercomputers and high end servers than for a competitors’ graphics cards that compete with its own line.
The Nvidia DGX H100 is an AI compute server, that uses the NVIDIA tensor cores, so it is not used for computer graphics. The previous version of this product used an AMD EPYC CPU.
Intel has had superior single thread performance for some time now so this is not a surprise.
It could be Intel made them a good deal on “mostly functional” Sapphire Rapids CPU’s:-) Given it is attached to the NVIDIA accelerator the accelerators built into SR are redundant.
Alan
Correct me if I’m wrong, but doesn’t that “primary reason” justifcation of “exceptional single-threaded performance” make zero sense in those supercomputer applications they’re talking about?
(ie building systems with 1000s of cores that are to be all used as efficiency as possible)
The compute will be done on the NVIDIA Tensor cores. The X86 is more of the supervisory and I/O agent.
It seems likely you could have your big array of tensor cores sitting idle waiting for instructions from the X86, so fast single thread on the X86 makes sense.
Alan
I agree with Alan. Each DGX has 2 CPUs. Their primary job is to keep 16,896 CUDA cores fully fed. AMD was just used for one generation when Intel was in a really bad spot.
I kinda expect Nvidia to make their own custom ARM to fill this role. But I suppose the volumes are too low to make it feasible.
I kinda expect Nvidia to make their own custom ARM to fill this role. But I suppose the volumes are too low to make it feasible.
If they were sufficiently determined to get around AMD and Intel they could go with one of the newer ARM server CPU vendors like Ampere, once that proves out. But it’s probably not worth the effort. Something custom… would have to deliver really specific benefits vs. just general “high single-thread performance.”
I think the threat that AMD is becoming to Nvidia is being overlooked.
AMD is taking a very large share of the Supercomputer, Server and HPC markets and looks to be accelerating that with the acquisitions of Xilinx and Pensando. Now add to this that AMD is starting to target Nvidia’s bread and butter in the AI space…
(The entire video is good but specifically watch the section starting at 14:43 titled “Why 2023 and 2024 Look horrible for Intel”)
Now let’s add to the above, Caromero1965’s announcement of AMD’s $1 billion in senior notes. I think this may be an indicator that Lisa Su is seizing an opportunity and stepping on the gas.
I would also call to your attention that if we compare the 12 month trailing revenue of AMD as of end of 1st quarter (March) to Nvidia’s annual revenue as reported in April; we see that in 2018 AMD generated 52% of the revenue that Nvidia generated. In March of 2022 AMD generated 82% of the 12 month trailing revenue Nvidia did for the 12 months ending in April.
I am sure that Nvidia is trying to figure out what to do about AMD. The move back to Intel may have something to do with that.
TSMC has been raising the price for leading edge wafers. Nvidia told them to pound sand and went with Samsung, only to end up in a bit of trouble. Rumors are floating about that the Samsung foundry is struggling even more than Intel. I suspect Nvidia has TSMC as their primary source and views Intel (rather than Samsung) as the backup.
Building on other comments, it appears the NVIDIA Cuda moat is eroding. We had a fairly long thread maybe a year ago about SYCL, and its impact on CUDA. Most technical people here felt CUDA would continue to be dominant, and it clearly still is. However, SYCL (and perhaps Intel oneAPI as a framework) is making significant progress at toppling that monopoly.
With both ARM, and accelerators at the door, I see why Intel is moving back to becoming mostly a silicon manufacturer.
Alan