Ask HN: Could advanced M-Line chips ever substitute for Nvidia H100s?

#1

Post by Devolver » Sun, May 12, 2024, 6:32 PM UTC

I'm researching the AI GPU market landscape...

I don't have the tech background in semiconductors to evaluate this question:

In theory, in (say) 2 years from now, could Mac Servers (with, say M5 Ultras) be used in a data center as a substitute for NVIDIA's H100s?

If so, why isn't anyone other than Apple exploring this direction?

It not, why not?

If you grok the tech, please explain to someone who doesn't. THANK YOU!!!

Re: Ask HN: Could advanced M-Line chips ever substitute for Nvidia H100s?

#2

Post by p1esk » Tue, May 14, 2024, 6:21 AM UTC

Nvidia has about a hundred direct competitors in AI accelerator space, both in datacenters and in edge devices. Apple is typically only interested in edge devices. Mac servers, despite the name, are edge devices. It is possible that Apple decides to make their own datacenter chips but I don’t really see it. It’s not their area of expertise. Even Amazon struggles pushing their AI chips, no one wants them.

Re: Ask HN: Could advanced M-Line chips ever substitute for Nvidia H100s?

#3

Post by uptownfunk » Tue, May 14, 2024, 7:06 PM UTC

Winners generally stick to what they’re good at. Apple is consumer hardware- everything else is just accessory. Nvidia is a whole company just focused on this one thing. The markets tend to favor this type of behavior.

Re: Ask HN: Could advanced M-Line chips ever substitute for Nvidia H100s?

#4

Post by fizzynut » Wed, May 15, 2024, 2:26 PM UTC

An M5 ultra would be the best ai chip for a data center in the same way it would be a good idea to use an array of the best laptop speakers to power a music concert.