I have a pair of H100s on a single machine. Nvidia-smi shows two GPUs with power cap of 310W. The spec sheet says 200-350 (configurable). But when I try to increase it;
~$ sudo nvidia-smi --power-limit=350
Provided power limit 350.00 W is not a valid power limit which should be between 200.00 W and 310.00 W for GPU 00000000:27:00.0
How can I increase it?
That spec sheet/specification covers all H100 (PCIE). It does not mean that every H100 can be configured to 350W. But some can. Whether or not you can achieve a particular TDP is also dependent on the system your GPU is plugged into. So the error message you have gotten is probably correct:
even though the “spec sheet” may say something different. You can confirm this by contacting your system OEM.
Thanks for the quick response!
I will follow up with our OEM.