News for hackers. Stuff that matters.
Re:Mistral Large 4 by amelius on Tuesday October 06, 2026 @10:26AM (#49979038)
> have not been a winner-take-all runaway acceleration game where catchup is impossible

From the Mistral site:

> ML4 was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral’s own datacenters in Europe.

It is pretty capital intensive!

Re:Mistral Large 4 by eigenspace on Tuesday October 06, 2026 @11:27AM (#49979965)
That cluster is literally orders of magnitude smaller than the compute pools used by Anthropic or OpenAI.
Re:Mistral Large 4 by amelius on Tuesday October 06, 2026 @12:39PM (#49980919)
For training or for inference?
Re:Mistral Large 4 by ricardobeat on Tuesday October 06, 2026 @01:21PM (#49981476)
They don't publish numbers, but Anthropic has a single DC with 200k+ GPUs for inference, GPT-6 Astra is said to have trained on 100k+ GPUs.
Re:Mistral Large 4 by anvuong on Tuesday October 06, 2026 @02:03PM (#49981932)
Both, especially for training. Astra and Fable were presumably trained on cluster of 100,000k GPUs, or at least a couple of 10Ks.

3,800 GPUs is nothing in the frontier side.

Re:Mistral Large 4 by dannyw on Tuesday October 06, 2026 @12:17PM (#49980647)
That’s kinda very small and light for modern trillion-param LLMs.
Re:Mistral Large 4 by bjenkins358 on Tuesday October 06, 2026 @10:45AM (#49979310)
I’m pretty impressed that they managed to get that close to the frontier with such a small cluster!
Re:Mistral Large 4 by locknitpicker on Tuesday October 06, 2026 @12:34PM (#49980838)
> I’m pretty impressed that they managed to get that close to the frontier with such a small cluster

Chinese companies also managed to put together their models with relatively small clusters.

Perhaps US companies are desperately trying to brute force their way into workable models?

Re:Mistral Large 4 by jayd16 on Tuesday October 06, 2026 @11:46AM (#49980236)
These cards are like $3k each? That's, what, $12M and you keep the hardware? Honestly doesn't seem too bad.
Re:Mistral Large 4 by amelius on Tuesday October 06, 2026 @12:01PM (#49980458)
More like $30k each.
Re:Mistral Large 4 by jayd16 on Tuesday October 06, 2026 @12:14PM (#49980623)
Oh the server chip is 10x. That makes a lot more sense.
Re:Mistral Large 4 by everfrustrated on Tuesday October 06, 2026 @10:43AM (#49979275)
According to Grok thats 7-10 MW. Tiny numbers.

To put that into context, the last wave of capacity SpaceXAI added 400-450 MW.

Re:Mistral Large 4 by amelius on Tuesday October 06, 2026 @11:03AM (#49979565)
But how much of that are they using for training versus inference? They're serving quite a large user base.