The gpu activity can be tracked with the following code segments:

nvidia-smi

Running nvidia will give useful information about your gpu such as driver version, how many processes are happening, GPU memory usage

nvidia-smi output

To see real time GPU tracking, call nvtop:

nvtop

nvtop will bring up a live graph to track the the gpu which gives good intel including memory and gpu usage.

For example running pet_breeds from the fastbook. Downloading the data uses 25% GPU usage

ISSUES:

A issue that can occur in this process is if the gpu reachs 100% memory the remote laptop will kick you off the server. If this happens just exit the server what a few minutes and sign in again.

GPU Full memory Error

The the server doesn’t kick you out and the GPU memory is full. An error like this might occur:

---------------------------------------------------------------------------
OutOfMemoryError                          Traceback (most recent call last)
Cell In[17], line 2
      1 learn = vision_learner(dls, resnet34, metrics=accuracy_multi)
----> 2 learn.fine_tune(3)

The solution to this is to simply restart the kernal which will get rid of any previously stored models.