We’re using the pyNVML Python interface to NVML for GPU device monitoring. The NVML C library contains an nvmlDeviceGetCudaComputeCapability() API for obtaining CUDA compute capability ratings, however this API isn’t exposed by pyNVML. Could pyNVML be be enhanced to support the nvmlDeviceGetCudaComputeCapability() function?
Yes, but NVidia does not update this library from 2015 (https://pypi.org/project/nvidia-ml-py/#history). Do it yourself in last sources :-) - https://pypi.org/project/nvidia-ml-py/#files.