Noteing the different ways of reporting the permormance, let me suggest this as a simple way. I counted the tempature humps in Efmer TThrottle and determinded the average completions per hour. This takes all factors into account. Here's what I Got.

GTX 285 (X58, I7 950 stock) 9.5/hr or 7.2 min/WU
GTX 570 (Z86, 2600K@4.4GHz) 20/hr or 5 min/WU

The GTX 285 was throwing about 40% errors until I reset the CPU to stock speed. I also got a little improvement when I when to MSI afterburner and set a very steep fan ramp to evoke 100% if 70C is reached.

This is great data and very helpful. Hopeing one of us will collect this in a spread sheet soon.

Based on all I've read here, I just ordered a 7970.

Now what about a 10, 20, & 50 year badge....Hmmm. Well thats for another thread :)
Thanks for the info fellow crunchers.

B2I

Very interesting. Hyperthreading off speeds up the GPU.

HD 7950 @ 1100MHz core & FX-8120 @ 4Ghz

8 work units simultaneously to complete one work unit every 30 seconds. Kernel time per single work unit is ~19 seconds.

I'd like to see someone with a HD 7970 clocked at 1300MHz turning in results, I know they are out there.

The old HT debate continues.
The statement that "two hyper threads equals one real core" is very wrong.
That being said if you disable HT your GPU tasks will run quicker and your will have less errors. True you can get "up to" 20% more work done doing CPU tasks but with my experiance it's not worth it when mixing GPU and CPU work units.

For the WCG CPU projects I run, I get almost 50% more work done on my Core i7-3770 (with HT) as compared to my Core i5-3550 (without HT). The speed of the i7 is slightly more, but not enough to account for more than maybe 5% of that. But I have only tried it on CPU projects, so don't know what HT does to the GPU projects in a mixed environment. I expect it depends a lot on which of the CPU projects you are running (e.g., CEP2 will probably be different than anything else).