I am very intrigued by the face recognition app, and I am looking into building a new machine anyway. I understand that a GPU with CUDA support is highly recommended, but I fail a bit in assessing how much grunt is really needed. I need some help in assessing what parts I should buy
I do not want to spend a large sum on money, so I am sourcing all components second hand. There seems to be a fair amount available, but I have no idea if the cheap (sub $50) GPUs are enough due to a complete lack of experience of these kinds of things. The ones I am looking at has 92 - 192 CUDA cores, and 1GB of RAM. Is this sufficient to accelerate performance to an “acceptable” level (max 10s per photo)? The wiki on Github suggest 2 - 4GB of RAM if CPU processing is used, but nothing for GPUs. Do I need a GPU with that amount of RAM, or is just “any” GPU enough to accelerate?
My server has only two users, but there is at least 100GB of photos to process. I am also looking into using this card for speech recognition once mycroft/mozilla deepspeech release it in a form usable for noobs like myself.