Faceswap Forum

Posted: **Tue Apr 07, 2020 9:36 am**

I'm not sure whether 8GB of GPU is enough or do I need something bigger?

Posted: **Tue Apr 07, 2020 10:24 am**

I have trained Villain at a very low batchsize (4 iirc) on a GTX 1080.

This was on Linux, so will be tougher with Windows, which holds back more of your GPU RAM.

You should be able to get it going with some of the VRAM saving options though.

Posted: **Tue Apr 07, 2020 10:37 am**

FYI

Windows, 32Gb
4 batch size with single GTX1070ti (8GB)
8 batch size with 2 x 1070ti's in SLI. (16GB)

Takes a long time !

Posted: **Tue Apr 07, 2020 10:54 am**

djandg wrote: ↑Tue Apr 07, 2020 10:37 am
Takes a long time !

I'll second that. Took me over 10 days to train Villain to convergence

Posted: **Sat Apr 11, 2020 3:17 pm**

How do you know when Villain has achieved convergence and can it handle 512px?

Posted: **Sat Apr 11, 2020 4:08 pm**

See the training guide. Specifically the section "Monitoring Training". viewtopic.php?f=6&t=146#monitor

No. It's a 128px model

What is the size of the GPU needed to train the Villain model?