Originally Posted by Mesmero
From a profiteering stand point wouldn't that be the dumbest thing Nvidia could do?
uhh not really?
The 590 ran out of production long time ago.. barely any shop has it in stock.
The HD 7970 is already performing pretty close to 590 / 6990 and if the GTX 680 turns out to be a few percent faster it might indeed tie the dual-GPU cards in performance, most likely it will OC better too.
The 480 was also on GTX 295 level or even faster, i dont see anything wrong with it.
Originally Posted by kweechy
Here's what I don't understand...the clock speed is nearly 30% higher than the 580 and it has 3x more CUDA cores, yet is still only ~30% faster than the 580. Wouldn't the overclock alone account for that?
when will people realize that you CANT compare the amount of CUDA cores / streaming processors between different architectures...
The HD 6970 had 384 (4D) shaders as opposed to the 7970's 2048 (1D) shaders, but is the latter 5 times faster? No.
The GTX 285 had 240 shaders whereas GTX 480 had twice as much, but was it twice as fast? No, not even close.
The new Kepler architecture appears to rely on a much higher count of weaker performing CUDA cores than Fermi that, taken together, appear to perform better in a way, be it raw computing power or maybe simply a more power efficient operation to cut down power consumption, simple as that.Edited by toX0rz - 3/19/12 at 2:46am