Deploying DeepSeek on 96 H100 GPUs

Re:Deploying DeepSeek on 96 H100 GPUs by fooker on Friday August 29, 2025 @09:14PM (#45071087)

The software stack for doing what you suggest would cost about a hundred million to develop over five-ten years.

[ Parent | on HN ]

Re:Deploying DeepSeek on 96 H100 GPUs by appreciatorBus on Saturday August 30, 2025 @01:56AM (#45072220)

But I was assured that this sort of stack could simply be vibed into existence?

[ Parent | on HN ]

Re:Deploying DeepSeek on 96 H100 GPUs by derefr on Monday September 01, 2025 @01:48PM (#45094927)

Yes, "HPC workload-scheduling software with multi-tenant customer usage accounting" does cost a hundred million dollars to develop and takes 5–10 years to build.

But some research labs (Lawrence Livermore National Laboratory, the research arm of HP, and a few others) got together to build it ~2002, and decided to make the results open source.

And that's what SLURM is. No, really.

> Slurm is the workload manager on about 60% of the TOP500 supercomputers.

[ Parent | on HN ]