Enable access for NVIDIA GPU profiling
Hey!
TL;DR: Can we enable access to GPU performance counters by default as described here?
I do research on GPU optimization and typically this entails profiling with NVIDIA Nsight Systems or Compute. Profiling (in-depth) today fails on Runpod pods due to this error.
The steps suggested there (
TL;DR: Can we enable access to GPU performance counters by default as described here?
I do research on GPU optimization and typically this entails profiling with NVIDIA Nsight Systems or Compute. Profiling (in-depth) today fails on Runpod pods due to this error.
The steps suggested there (
.conf file + update-initramfs ...) do not work on pods, at least that has been my experience. Can we have a template that has this configuration set by default?