Does anyone have metrics for some sort of benchmark per transistor? I wonder where the sweetspot is between something like these and SERV [1]. Obviously GPU win

Does anyone have metrics for some sort of benchmark per transistor? I wonder where the sweetspot is between something like these and SERV [1]. Obviously GPU wins on parallel math, but I wonder how much we are chasing in terms diminishing returns. [1] https://www.youtube.com/watch?v=GSHasXHvZaQ https://www.youtube.com/watch?v=GSHasXHvZaQ