NVIDIA CEO Isn’t Intimidated by AMD’s New Multi-Die GPU, Laughs at the Concept of an “NVIDIA Killer”

NVIDIA CEO Jensen Huang didn’t seem worried about AMD’s recently announced new multi-die GPU: “We don’t feel like we need to compete with them. We know our strengths and when you have a strength, there is no need to worry.”

NVIDIA CEO, Jen-Hsun Huang, isn’t intimidated by AMD’s new multi-die GPU. He laughs at the concept of an “NVIDIA killer.”

NVIDIA image

AMD recently unveiled its Instinct MI200 Series Accelerators, HPC GPUs that are garnering a lot of excitement for being the world’s first multi-die GPUs capable of a 4.9x performance boost over competitor devices, according to AMD. Despite the achievement’s promise of up to 47.9 TFLOPS of FP64 performance and 3.2 TB/s of memory bandwidth, NVIDIA CEO Jensen Huang is unconcerned. Huang shook his shoulders when asked about the development by Tim Morgan of TheNextPlatform in a recent interview, implying that the new Instinct was simply another alleged “NVIDIA killer” that would fail to live up to that boast.

Morgan’s discussion with Huang about NVIDIA’s competitors, software, and the Omniverse includes the following excerpt:

Timothy Prickett Morgan: And now, for GPUs, DPUs, and, shortly, CPUs, there is a two-year cycle that organizations can rely on.

AMD’s “Aldebaran” Instinct MI200 series GPU accelerators are clearly more competitive than they have ever been. Although it is two GPUs, not one, and I reminded everyone that AMD had “pulled a K80” by placing two GPUs on one chip, this GPU has won two exascale-class systems and a slew of other smaller systems.

Given the cycle, I understand that Nvidia will not make any new GPU announcements until next year, but what is your reaction to AMD’s and, to a lesser degree, Intel’s competitiveness in the GPU computing arena?

Jensen Huang (Hong Kong): First and foremost, we are always competing. So it’s not accurate that this is the first “Nvidia killer” to appear. Every year, there is an Nvidia killer, as it is known.

The NVIDIA CEO goes on to say that competing in this market isn’t as difficult as it seems. Huang seems to believe that improving performance and capacity while maintaining competitive price is all that is required.

“[…] I believe this is the most straightforward area, and let me explain why,” Huang added. “This is because an HPL machine just requires two things – just two things.” HPC centers purchase everything years in advance, so they have no clue what the performance of any specific device will be, but here’s the equation for you.”

“Put the amount of peak FP64 flops and the memory capacity into the same bucket. Put money in one bucket and dollars in the other. That is all there is to it. That’s the formula.”

However, AMD’s Instinct MI250X accelerators have already won a big contract with the US Department of Energy, Oak Ridge National Laboratory, and HPE’s newest supercomputer. The supercomputer, dubbed Frontier, will have a peak computational capacity of more than 1.5 exaflops courtesy to AMD’s latest GPUs and optimized 3rd Gen EPYC CPUs.

TheNextPlatform is the source of this information.