And for the industrial market $ per performance is all that matters because in large deployments there is no issue with just parallelizing as many GPUs as you want. Even if an intel GPU for a 10th of the price has a 5th of the performance, then you just slap together 5 of them and get the same processing power for half the price.