David.Podedworny wrote:
Mad Ad wrote:
I didnt know Nvidia had any fabs to compete head on with the others, gpu design and construction typically lags behind a generation or two in process size, if they try making cpus on old technology theyll end up down the same alley as cyrix
Actually no it should be up to par one scalar partition(16 brains or 16 pipeline) from a g80 is about equal to a core 2 e6600 in simple math operations.
plus the design is all wrong, gpus are good at parallel math, hence the assumption that spare gpu power could be easily offloaded to process physics (rather than a discrete pys card), another parallelised flow however until we start to get fully multithreaded kernels then a cpu seems best designed as a single pipeline- plus all the traditional thinking in programming is geared towards set extensions and the traditional model, even if a massivly parallel o/s was to flop out of the ether it would take generations to learn the new approach.
No I firmly belive that an Nvidia cpu would be dead in the water if they try to compete with AMD/INTEL head on without their own fabs to drive the shrink too, theyd be too reliant on the spare capacity of 3rd party manufacturers and always be playing catchup on every front.