open-mmlab/mmdetection

I used a single server with 8 RTX 3090 GPUs to train GLIP, which took 4 days. However, using two servers with 16 GPUs took 9 days. Is this normal?

CLL112 opened this issue · 1 comments

CLL112 commented
I used a single server with 8 RTX 3090 GPUs to train GLIP, which took 4 days. However, using two servers with 16 GPUs took 9 days. Is this normal?
CLL112 commented

Is it possible that the network connection between my servers is causing the slow performance, or is this speed normal?