In jetsons, you have the ability to change the number of CPUs that are active, for example using all CPUs or only 2 CPUs.
When training a machine learning model I would always use the GPU cores for computations. My question is, since I am transferring over all the computations for my model to the GPU, is there even a point in changing the active number of CPUs?
If yes, How does changing the active number of CPUs affect your GPU computations for machine learning applications?
Does changing the active number of CPUs affect your GPU computations for machine learning applications?
No. It won’t have too much difference if all the tasks are deployed on the GPU.
But it’s recommended to pick a mode with higher clock rate. Ex.
nvpmodel -m 0
This can have an optimal GPU launch and synchronized time.