Closed DonggeunYu closed 5 years ago
fix it!
Hi, sorry to bother you. May I ask how did you fix it? I added CUDA_VISIBLE_DEVICES to be 0,1, now the code occupies both of my GPUs but my training speed remains the same as in training with single GPU.
I'm seeing the same issue that it is only using 1 of my available gpus. It starts multiple processes for each card but nvidia-smi shows that all the processes are using the same card. I also see no increase in performance from additional cards.
I used TITAN XP x2. I tried:
but Errorororororororor. Help me please!