Nvidia Fermi performance, frequency vs #cores

[Blasko9]

Maybe a dumb question but....

I have two GF110s in a single box, a GTX580 @797mhz and a GTX560ti 448 @797mhz. The 560ti has 12.5% fewer cores so if I can get a stable overclock @910mhz, should I expect to get the same folding performance from both cards?

It does not appear that way because @ stock freqs the GTX580 is about 30-40% faster on the same WU. Do more cores running slower out pace fewer cores running faster?

[7im]

More cores is always better. Fire hose vs garden hose. Even on low power the fire hose will knock you over.

[uncle fuzzy]

In general, yes.

More cores- good.
More cores running faster- more good.

[bruce]

What I normally do is multiply the two numbers together to create a figure of merit, provided they're the same architecture. (Twice as many cores can run a about half the speed to get the same amount of work done.) It's not a perfect measurement by any means, but it's closer than only looking at either of the numbers in isolation.

Comparing the two at stock speeds:
GTX560ti: 448 @ 732 ---> 328
GTX580: 512 @ 772 ---> 395
Ratio: 1.20

Comparing the theoretical GFLOPS:
GTX560ti: 1263.4
GTX580: 1581.1
Ratio: 1.25

I'll let you figure it out for your clock rates. (and by the way, theoretical GFLOPS is not an especially accurate prediction of FAH throughput either, but it's better than nothing.)
Do your own benchmarking.

[EXT64]

Is that 30-40% faster TPF, or 30-40% more PPD (if you are doing WU with QRB, Core 17 WU, then this matters).

[Blasko9]

As I recall, both cards were working on a core 15 (14,000 points). The 580 TPF was 6:50 and the 560ti 448 TPF was 11:10. I was a bigger difference than I expected for 512 cores vs 448.

[artoar_11]

Just for comparison. My GTX 460 @ 775 MHz (336 CUDA cores) on p7620-7627 (core_15), TPF ~ 00:09:12 - 00:09:30. NV driver - 327.23 or older.