NVIDIA CEO Brushes Off AMD’s MI200 MCM GPU: Every year there is an Nvidia killer and people call it that


AMD not too long ago introduced its Intuition MI200 “Aldebaran” knowledge middle GPU, providing almost 5x extra FP64 throughput in comparison with NVIDIA’s prime A100 Tensor Core Accelerator. This, in fact, means NVIDIA must react to the event in a technique or one other. In an interview with TNP, NVIDIA CEO Jensen Huang dismissed the Intuition MI200 as simply one other competitor within the knowledge middle and HPC market.

Timothy Prickett Morgan: Now there’s a two-year cadence for GPUs, DPUs and shortly additionally for CPUs that firms can depend on.

Clearly, AMD is far more aggressive with its Intuition MI200 “Aldebaran” sequence accelerators than it has ever been. It’s really two GPUs, not one, and I’ve reminded everybody that AMD had “made a K80” by placing two GPUs in a single machine, however nonetheless, this GPU has gained two ex-scale class programs and lots of extra small programs.

I’m conscious that there will probably be no new GPU bulletins from Nvidia till subsequent 12 months, based mostly on cadence, however what’s your response to this competitors from AMD, and shortly, to a lesser extent, Intel within the area of GPU computing?

Jensen Huang: Initially, we now have competitors on a regular basis. So it isn’t true that that is the primary “Nvidia killer” to hit the market. Yearly there may be an Nvidia killer and folks name it that .

Timothy Prickett Morgan: I’m speaking in regards to the high-end AI and HPC supercomputer area. For the previous decade and a half, relating to GPU-accelerated supercomputers, you’ve been the one participant.

Jensen Huang: Truly, I feel that is the simplest area, and let me let you know why. The reason being that an HPC machine wants two issues, solely two. HPC facilities order all the things years upfront, in order that they don’t know what the efficiency of a given machine will probably be, however right here’s the equation for you …

Timothy Prickett Morgan: Go forward …

Jensen Huang: The variety of most FP64 FLOPs and the reminiscence capability, put these two issues in a dice. And within the different bucket, put {dollars}. That’s all. That’s the equation. And you already know that …

Timothy Prickett Morgan: And so AMD determined to extend FLOPs and reduce the worth? That’s what I feel occurred …

Jensen Huang: The query is how we see the world, and why the competitors is so intense for us. And it is extremely intense. It’s not of much less depth. It’s critically intense. Accelerated computing just isn’t for folks with a weak coronary heart. So let me show it.

You’ll be able to construct the most effective chip on the planet for all the things, you set it within the pc, and what are you going to speed up? Completely nothing. It’s not true? Accelerated computing is extremely tough. And the reason being that Moore’s Regulation is extremely good. Nobody has checked out Moore’s Regulation, even at its sluggish tempo, and mentioned over time that it isn’t one of the formidable technological forces in human historical past. And but, for us to achieve success as an organization, we now have to ship outcomes nicely above Moore’s regulation.

The Subsequent Platform (By way of: EC)

Stating that there’s competitors on a regular basis, Huang defined that the MI200 isn’t the primary “NVIDIA killer” to hit the market. He went on to say that there’s an NVIDIA killer yearly and folks name it that. Staff Inexperienced’s CEO believes that the high-end AI and HPC area is the simplest, requiring “simply” excessive FP64 compute efficiency and loads of reminiscence and bandwidth.

This image has an empty alt attribute; its file name is FDr0bCGWEAAgA1G-1-1024x582.jpg

Nevertheless, if we have a look at the capabilities of the MI200, it’s method forward of the A100 each by way of FP64 matrix and vector compute, whereas additionally providing a good bit of extra reminiscence and bandwidth. Add to that the truth that there gained’t be any new choices from NVIDIA within the knowledge middle marketplace for at the very least one other 12 months, and AMD has received a considerable benefit for a considerable period of time.

Be the first to comment

Leave a Reply

Your email address will not be published.


*