OOM with 11GB of ram

[andreas128]

How large does my GPU have to be to run the SRGAN?

[zsdonghao]

1080 should be enough

Sent from Netease Mail On 04/18/2018 17:58, andreas wrote:

How large does my GPU have to be to run the SRGAN?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[PeeJay]

I have the same problem. The first part of the training succeeds, the second part can't allocate enough memory. On Linux it will keep training regardless, but on windows it stops with an error. This was using a 1080ti with 11gb ram

[Obliviously]

Im not sure how much VRAM you need but you could change the code to load images only batch wise to reduce memory load. Apart from that you could lower the batch size or lower the image resolution. Last but not least you could also train on CPU which is obviously a lot slower.

[William-Mou]

Hello, thanks for publishing the code! I meet the same problem like @PeeJay. I was using 1080ti with 11gb ram on windows 10. I have been train the model to the second part in the past (about 1100/2000 epoach). But magically, when I restart and try to train it again, It stops with the error (OOM). In the same env, with the same code... So crazy!

[PeeJay]

FYI: I solved the OOM by reducing the batch size from 16 to 14