04-10-2018, 10:23 AM
(02-02-2018, 07:13 PM)FPTHHrw6mKhNZwL6 Wrote: Is it about stream processors? Then that would make AMD win the race...?
There are more factors. ALUs and core clock are one side, but how efficient is the code?
If you have GPU 1 with 1000 core and 1GHz clock rate, and GPU 2 with 1200 cores and 1.25GHz clock rate, then the second would be theoretically faster.
But if a code routine of GPU 1 needs only 100 cycles and with GPU 2 it needs 150, then their speed level is at par.
This is possible because the code is operated in small steps and each step needs 1 to x cycles. If a 64bit-Division needs 3 cycles at GPU 2 and 2 cycles at GPU 1, the latter one needs 1/3 less time.
And then there can be also different instruction sets, with faster instructions coded into the hardware-design.
NVidia is leading especially in code efficiency, that's why they take the lead over AMD.