News for hackers. Stuff that matters.

A comment on Nvidia H100 and A100 GPUs – comparing available capacity at GPU cloud providers. Parent comment

Re:Nvidia H100 and A100 GPUs – comparing available capacity ... by sbierwagen on Wednesday June 14, 2023 @08:23PM (#36334813)
>heavily differentiating the H100 products by their interface this go around, which I find quite odd.

Power draw. The 4090 pulls 450 watts, and is destroying connectors. H100 SXM uses 700 watts.

If the PCIe consortium wants NVIDIA to use their standard for top end products, they need to make a better connector.

[ Parent | on HN ]
Re:Nvidia H100 and A100 GPUs – comparing available capacity ... by lyu07282 on Thursday June 15, 2023 @06:50AM (#36338673)
It was a cheap power adapter design that was sold by Nvidia that caused the melting, it was nvidias own doing selling a cheap cable on a $1,600 card. Using proprietary interfaces has everything to do with monopolistic behavior and nothing with faulty standards.
[ Parent | on HN ]
Re:Nvidia H100 and A100 GPUs – comparing available capacity ... by nullindividual on Friday June 16, 2023 @05:43AM (#36353917)
12VHPWR is art of the ATX 3.0 standard and not proprietary to nVidia.
[ Parent | on HN ]
Re:Nvidia H100 and A100 GPUs – comparing available capacity ... by lyu07282 on Friday June 16, 2023 @04:05PM (#36363066)
The adapter caused the issue not the connector itself, nobody who used power supplies with a native connector had any issues.

I was talking about the SXM interface instead of PCIe

[ Parent | on HN ]