iperov
gpu not responding when extracting facets
Expected behavior
trying to use data_src facest extract.bat, would expect it to find faces as it did with my older gpu (gtx 1080)
Actual behavior
takes a long time displaying the 'extracting faces...' message, then shows the progress bar which dosent go up and after a few minutes it displays this message 'eForce RTX 3090 doesnt response, terminating it.' saying all of the images were found but no faces detected
Steps to reproduce
only thing i have changed is the graphics card. had a gtx1080, now have a 3090.
Other relevant information
windows version 10.0.19042
system specs: 5950x 16 core processor 32gb ddr4 3600 ram rtx 3090
also using a riser cable but it seems to be working fine for everything else
I have the same problem with my 3080 with the latest nvidia DFL build. I tried reinstalling tensorflow and cuda11.1... Did you find a way to make it work ?
I am having the exact same issue. Using 2.0 . I tried DDU with full driver updates for GPU, complete clean wipe and reinstallation of windows, and nothing has helped. Ran 3d Mark to test the GPU was okay and it worked fine, have my HDMI plugged directly into GPU port and everything and it works fine. I'm not sure what the issue is but it seems to be deepfacelab specific. Weird thing is like 3 weeks ago my GE Force RTX 3090 worked just fine.
You CAN in the meantime have your facesets extracted with CPU, but that turns what my GPU could do in 1 hour to a 36 hour task which is not going to work for me personally. Very interested if anyone has found any solutions?
I have the same problem with my 3080 with the latest nvidia DFL build. I tried reinstalling tensorflow and cuda11.1... Did you find a way to make it work ?
Did reinstalling tensor/cuda work for you?
I have the same problem with my 3080 with the latest nvidia DFL build. I tried reinstalling tensorflow and cuda11.1... Did you find a way to make it work ?
Did reinstalling tensor/cuda work for you?
No, I think it solved itself by trying multiple times and at some point the extraction just started but veeery slowly. and training was using mostly CPU (even if i selected GPU) and the GPU Vram at the same time.
Does tensorflow import in your environment well? I had that problem and found out that tf couldn't import properly. Try it manually in your env.
did you find a resolve to this issue, i just got a nvidia rtx 3060 and im having the same issue it looks like
Does tensorflow import in your environment well? I had that problem and found out that tf couldn't import properly. Try it manually in your env.
My cuda version is 11.x and in tensorflow, "tf.test.is_gpu_available()" gives me a 'True'. But I cannot extract faces using 3090 GPU, just the same as tomavisar mentioned. And it turns out that ONLY face extraction using S3FD cannot run on GPU successfully.
Also having this exact problem with my nvidia RTX a2000. my deepfacelive install works fine. I have tried tinkering with CUDA versions and tensorflow, but actually couldn't manage to install tensorflow 2.10 - it would only let me install 2.12 which isn't CUDA compatible.
Any suggestions for GPU diagnostics from within DFL would be appreciated.
