Report me if I continue to console war
- Feb 22, 2017
The example cerny gave resulted in the 2 gpus equaling the same tflop performance
So instead of this:
48cus @ 1674mhz = 10.28tflops
Using 36cus @2230mhz = 10.28tflops
Tflops is performance of the vector ALU's , however at higher clockspeeds rasterization performance is higher in and command buffer processing is also faster inline with the clockspeed increase, also L2 and other caches have higher bandwidth at higher clock speed
So while there are benefits in clocking less cus very high, its not completely a free ride the higher the clock the further away the mem is in terms of cycling, so while cerny said the benefits far outweigh the negatives, he is talking about 2 different approaches which result in the same tflops.
Also when comparing the ps5 gpu to the XSX gpu an advantage the xsx has is that it is at a constant performance level, there are no variables like the PS5, we also dont know how the PS5s added 22% clockspeed over the xsx will impact actual performance, we really need a game developer to flat out tell us the performance differences, my gut tells me that xsx GPU will still be about 20% more powerful when you factor in the 25% extra ram bandwidth it has.
I would also be interested to see benchmarks done on a 22% lower clocked GPU with the same tflops as one clocked 22% faster.