We all know that the Nvidia-smi returns the percent of used time over the past sample period, but it cannot reflect how much the GPU is used, such as how many tensor cores are used.
Are there any ways or tools to check the utilization of a process?
We all know that the Nvidia-smi returns the percent of used time over the past sample period, but it cannot reflect how much the GPU is used, such as how many tensor cores are used.
Are there any ways or tools to check the utilization of a process?