Dear Nvidia developers,
Recently, I switched from Cuda 10 to Cuda 11 and I noticed that some functions in NVML are marked as deprecated, but no further explanation is given what should be used instead.
More specifically, I used a script to read the NV-link counters like such:
./EXE
nvidia-smi nvlink -gt d
nvidia-smi nvlink -re #ERROR, doesn’t work in CUDA11
What is the preferred way in CUDA 11 to read out NV-link utilization after a job?
Thank you!
Best regards