Faceswap Forum

I was not sure where to put this, as it is persistent in all phases, so forgive me if this does not belong here.
I have 2 AMD Radeon Vega 64 GPUs, and they are both properly connected and are both detected by FaceSwap in both logs and Global Options. However, whenever I extract, train, convert or do anything, it is very apparent in both AMD Adrenaline and MSI Afterburner that only one of them has VRAM being utilized by FaceSwap. They are both fully functional since I can use Exclude GPU to switch between which one is being utilized and each one gives about 8GB of VRAM when being utilized. Does anyone know of a way to fix this so both are being utilized?

At this moment in time the DirectML plugin does not support multi-gpu. Only Nvidia does with Cuda.

I tried to find where I read this, but cannot find it in their Github. It may have been written in the source code.

Unfortunately, this is a current limitation of the plugin. Hopefully it will be resolved at some point in the future, but it is out of our control.