0

I have a three connected GTX 1080 Ti and I am trying to train a model with multigpu option. But it doesnt work. With single gpu, it train well but with multigpu it says like this and doing nothing:

I1210 13:24:25.852613  3033 net.cpp:1358] [1] Reserving 582912 bytes of shared learnable space
I1210 13:24:25.852761  3032 net.cpp:1358] [0] Reserving 582912 bytes of shared learnable space
I1210 13:24:25.852787  3034 net.cpp:1358] [2] Reserving 582912 bytes of shared learnable space

I have waited long enough then I have to kill the process. It uses 100% of all of the gpus after this command. What should I do?

Furkan
  • 3
  • 1

1 Answers1

0

In AMD systems, while trying to setup Ubuntu, you need to bypass GPUs.If you try to train a model with this selection, you get this error.

sudo nano /etc/defaults/grub

GRUB_CMDLINE_LINUX="iomnu=soft"

sudo update grub

dorukoski
  • 138