GPU utilization in Pytorch 0.4

When I specify to run my code on specific GPU(GPU-2), it uses that GPU specified but also creates subprocesses of around 500MB in GPU-0.

Assigned GPU using:

network.cuda(2)
image.cuda(2)

Moved data to GPU using:

mydata.to(2)

Pytorch version used: 0.4

Assigned GPU: Green (GPU-2)
Subprocesses: Red (GPU-0)

Why does this subprocesses PID 33106 GPU-0 get created?

1 Like