Warning on Ampere (30xx series) cards.

[bryanlyon]

[mod]At this time, 30xx series support is available, but you must update to the latest following the information in viewtopic.php?f=4&t=1226 [/mod]

Right now it is not possible to use Faceswap on these cards as they're incompatible with existing libraries. We're aware of the situation and are paying close attention to it. We'll support the cards as soon as the upstream projects have added support.

Feel free to "Subscribe" to this thread as we'll update it if/when there is any news on support for the new cards.

[bryanlyon]

Tensorflow has released 2.4 RC1. This is a good first step toward support. However, we're still waiting on the full version to be released and deployed to pip/Conda.

[bryanlyon]

Tensorflow 2.4 final has been released. Expect it to be built and added to Conda/Pip soon.

HOWEVER, it's important to note that this generally takes weeks to months to get released for Windows. Linux is expected to come first. If you have a 30xx card and want to use it under Windows, you'll have a wait, compile it yourself, or find a wheel. Compiling/getting custom wheels is at your own risk and is NOT supported.

[bryanlyon]

There is now an UNTESTED guide on using an RTX 30xx series card with Faceswap at viewtopic.php?t=1226 . It's very important to note that this is NOT TESTED and is likely to not work.

[torzdf]

Cuda 11.2/cuDNN 8.1 support has now been added to the default installer, so 30xx cards should just work.

See this thread for updating: viewtopic.php?f=4&t=1226