i see that a TPU is much more efficient to use than a gpu (cuda cores) and in some cases faster.
i'm currently running 35 ip cams on blueiris with codeproject AI YOLO5 cuda with gtx 1060.
i came across coral TPUs and yes they are more efficient compared to gpu cards.
Problem is with scaling, i cannot use multiple tpu cards on codeproject.
it seems like the rtx 2060 is the cheapest TOPS/$ on int8.
question is, can i use something like an rtx 2060 super which got 119 TOPS int8 for codeprojectAI without using the cuda cores (only vram and tensor cores)
i'm currently running 35 ip cams on blueiris with codeproject AI YOLO5 cuda with gtx 1060.
i came across coral TPUs and yes they are more efficient compared to gpu cards.
Problem is with scaling, i cannot use multiple tpu cards on codeproject.
it seems like the rtx 2060 is the cheapest TOPS/$ on int8.
question is, can i use something like an rtx 2060 super which got 119 TOPS int8 for codeprojectAI without using the cuda cores (only vram and tensor cores)