I'd like to build a 2nd PC for faceswap so as to not monopolize my main rig while training, but money is kinda tight and I have a 13 yr old PC in my basement (AMD Phenom x4 955, 8 gb ddr3 ram, 240 gb ssd), if I were to simply put a decent GPU in there, perhaps a GTX 1080 which seem to go for around $100 on ebay these days, would this make a decent faceswap PC? I suppose the 8 gb of ram would make sorting slower, and the old cpu will make extracting and converting slower than a modern PC, but training (which consumes the vast majority of my time anyways) should still be decent, right? Before I buy a GTX 1080 I just wanted to run this idea by others to see what they think?
Talk about Hardware used for Deep Learning
It should be ok, but I make no guarantees. The data augmentation pipeline for training the model is all CPU bound, so this may bottleneck things slightly, however it is fairly well optimized.
I Dev on some fairly old/fairly slow Xeon processors and it seems to go OK.
My word is final