Skip to content
This repository has been archived by the owner on Nov 9, 2023. It is now read-only.

What's the best way to run it with the RTX 3k series? #5227

Open
cm7park opened this issue Jan 1, 2021 · 5 comments
Open

What's the best way to run it with the RTX 3k series? #5227

cm7park opened this issue Jan 1, 2021 · 5 comments

Comments

@cm7park
Copy link

cm7park commented Jan 1, 2021

GPU: Geforce RTX 3080 10GB.
Tensorflow-gpu : 2.4.0
Driver : 460.89
CUDA : 11.1.1
CUDNN : 11.1 (v8.0.5.39)
Operating system : Windows 10

image

Hello.
I tried running the 2080TI version that was recently provided. However, it does not proceed from step 4.

There seems to be an issue with the RTX 3080 right now. As a result of the search, there were packages related to RTX 3k around September. Is it no longer available? Are you preparing for an upgrade?
DeepFace still failed to launch due to GPU issues. We hope the RTX 3080 version will be released soon.
Or let me know if there is a good way.

Best wishes for a Happy New Year. Thank you.

@cm7park
Copy link
Author

cm7park commented Jan 1, 2021

Extracting faces...
Running on GeForce RTX 3080
0%|2 | 20/7217 [00:32<3:13:18, 1.61s/it]

It works, but it doesn't use the GPU...

@robustini
Copy link

With RTX 3070 no problems here.

@Joe-121
Copy link

Joe-121 commented Jan 9, 2021

On Linux, RTX 3090 works fine to extract faces. Same core python scripts and conda packages are used in Linux vs Windows, so there should be a way to update your files. However, I don't think you even need to do that: latest windows version that you should be using does not say "2080Ti version". It is "DeepFaceLab_NVIDIA_build_01_04_2021.exe". Check latest files on Mega or on the torrent magnet link in readme file

@sergiobd
Copy link

@robustini are you on windows?

@robustini
Copy link

@robustini are you on windows?

Windows and Linux, work on both.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants