Is there any way to split the nvidia-smi dashboard so that it shows the GPUs individually? The dashboard is great, but because it assumes you only have one GPU, you get amusing (but incorrect) numbers since it adds all GPUs together. For instance, in one of our 8 GPU servers, the temp is currently 252 degrees celsius and the fan speed is 327%. Going into the server, you can see that that is not the case as it’s split across 8 GPUs.
Hi, @UCBKurt. That is possible when
- using go.d/nvidia_smi instead of the python version.
- change
group_by
of the chart fromdimension
togpu
.
Alternatively, you can change group_by
to instance
(no need to switch to go version).