I remember when Tesla was building its own inference chips, and after about 2 years and billions spent, the whole effort was scuttled b/c they simply could not keep up with the iteration and R&D cycles of dedicated chip companies. I suspect the same will be the case with OpenAI vs Cerebras + Nvidia/Groq
Hes got a press release.
The issue is, baking something to silicon requires discipline and about 2 years.
This isn't something you can just change your mind on halfway through. Trust me, I know. You need a clear vision of what you want to support, why and what bits of a chip you need to achieve that.
Man, if only someone made like, chips that could lots of different calculations all at the same time!
Even in nvidia land rubin + LPU does a similar thing.
It has its downsides of course - if your traffic swings prefill heavy to decode heavy, you can't suddenly use your lpu for prefill. With GPUs they're totally interchangeable. Tradeoffs.
There is also the other idea where you run your attention layer on the GPU/TPU/Trainium and the FFN on the SRAM accelerator. Because KV cache is more difficult on cerebras etc, while MOE latency is easier to deal with
That sounds quite like...nonsense?
Chip companies work on years-long cycles. They know today what are they launching 4-5 years from now.