reframe
reframe copied to clipboard
Auto-detect GPU devices
This is also mentioned in https://github.com/archspec/archspec/issues/25 but we could start working on a solution for ReFrame already.
I was thinking about this as a part of #2276. As discussed in the archspec issue, we can use nvidia-smi to get a plethora of details about NVIDIA GPUs. For the AMD ones, there is an equivalent tool with python bindings. Besides detecting, if we could provide available memory, compute capabilities, threshold temperatures, etc., that would be quite useful!
There is also this: https://pypi.org/project/nvidia-ml-py/