News for hackers. Stuff that matters.

A comment on Cerebras Release 2.0: 50% Faster Training, PyTorch 2.0, Diffusion Transformers. Parent comment

Re:Cerebras Release 2.0: 50% Faster Training, PyTorch 2.0, D... by wmf on Wednesday November 29, 2023 @04:09PM (#38465360)
It sounds like they just recently got weight streaming working well which was my point: if you bought a CS-2 when it first came out you couldn't really use the off-chip memory and the on-chip memory wasn't enough to run LLMs.
[ Parent | on HN ]
Re:Cerebras Release 2.0: 50% Faster Training, PyTorch 2.0, D... by efrank02 on Wednesday November 29, 2023 @08:53PM (#38468272)
Interesting! But you think now that this is solved cerebras will be adopted more widely?
[ Parent | on HN ]
Re:Cerebras Release 2.0: 50% Faster Training, PyTorch 2.0, D... by wmf on Wednesday November 29, 2023 @09:09PM (#38468442)
It's hard to say. We'll probably never have good information due to NDAs.
[ Parent | on HN ]
Re:Cerebras Release 2.0: 50% Faster Training, PyTorch 2.0, D... by rbanffy on Thursday November 30, 2023 @11:06AM (#38475174)
> It sounds like they just recently got weight streaming working well

Even if it was working poorly, the CS2 is still a lot of computer. The question is whether it was price-competitive with Nvidia at the time for the workloads it was acquired for.

Cerebras offers them in a batch-processing cloud-ish model, so their prices should reflect utility to some degree.

[ Parent | on HN ]