I was wondering if anyone is facing the problem of low GPU utilization. It seems like the problem is stemming from using open_image function as it is quite slow. As a result, the usage of my GPU is usually 0% then it spikes up to 100% when it is training. Has anyone experienced this, or have a solution to this, or maybe it is a computer limitation?
Do you store the data on a SSD? For me, I encounter your problem if I train on data, which is stored on a hard drive disk. When I train on data, that is stored on a Solid State Disk the GPU is fully utilized.
1 Like
You sir are a genius!