Faceswap Forum

Hi guys, sry for disturb, I'd like to buy some 2nd-hand graph cards for faceswap, what's kind of model is better for faceswap?

See here:
viewtopic.php?t=10

Ultimately, buy Nvidia.

Hello,
I am fairly new to this but I have some information that maybe helpful.
It seems like NVidia is the way to go, but the there is a factor called "GPU Compute Capability".
At this website, you can find the score for the various cards.
https://developer.nvidia.com/cuda-gpus
The section about CUDA-Enabled GeForce and TITAN Products will have info for "gamer" cards like GTX 1080 through RTX 3090 TI.

I had recently purchased a GTX1070 for $170.00 and its score is 6.1. This causes a warning in FaceSwap's UI about the Swap taking longer but it works OK. What I found odd about the scores is that the 1070 and 1080 have same score and the beast card RTX3090 TI is 8.6.

So, for over $1000.00 for a RTX3090TI or $170 for the 1070. I chose the 1070.

Regardless - this takes a VERY long time to do.
I hope this helps!

Just an fyi: the compute capability isn't a score, but it is a version. The 10xx cards are version 6.1
The 20xx cards are version 7.5.

The main difference between 6.1 and 7.5 is that 7.5 added support for Tensorcores (that is the ability to train quickly using Mixed Precision) which explains the warning you are getting.

Basically, as new versions of compute capability are released, more features will become available that are not available on older cards.

A full list of capabilities can be found here:
https://en.wikipedia.org/wiki/CUDA#GPUs_supported

torzdf wrote: ↑Sun Nov 06, 2022 12:42 pm

Just an fyi: the compute capability isn't a score, but it is a version. The 10xx cards are version 6.1
The 20xx cards are version 7.5.

The main difference between 6.1 and 7.5 is that 7.5 added support for Tensorcores (that is the ability to train quickly using Mixed Precision) which explains the warning you are getting.

Basically, as new versions of compute capability are released, more features will become available that are not available on older cards.

A full list of capabilities can be found here:
https://en.wikipedia.org/wiki/CUDA#GPUs_supported

Hi,

The gtx1650 for laptops is not mentioned anywhere here
viewtopic.php?t=10
https://en.wikipedia.org/wiki/CUDA#GPUs_supported

Can it be used just to deepfake/swap normal quality images?

GTX1650 is listed there under compute capability 7.5