Hi, Chris:

I like your Big bang Theory!!! I did not say, I believed it, just liked it....

It's actually a rather hard comparison to make. It's definately faster than all gpu's when running FP32 tasks (sieves),

580 overclocked to 860MHz = 304s
gtx680 stock = 265s.

Someone also posted that a 680 is 44% faster than 570.

Einstein recently allowed 300+ drivers running, since they had blacklisted them due to monitor sleep bug. However, since their tasks require so much CPU work, all high-end GPU (560 448 cores and above) return work in about the same time (about 1500 sec)http://einstein.phys.uwm.edu/forum_thread.php?id=9097 . The difference ont their site seems to come from loading multiple WU on one GPU. Here it appears the 680 excels (these results are not mine)

On a 3930K @4.4= (3 tasks per GPU)

2604 sec
2610 sec
2559 sec

On his 580 in 920 @4.2: 3836 Sec Avg (850 MHz)with 3 running.
He said he was going to attempt 4, but has not posted again yet.

GPUgrid is the one where I think the best bench is going to come from, because they are currently creating a new CUDA4.2 app. I have not been able to run there (they currently error out). Gianni said he should hopefully have an app running towards beginning of next week, with a 680 optimized app out later, and I think this site will benefit the most, since the WUs can run so long, and will HOPEFULLY be able to take advantage of its CC 3.0.

The big jump on Einstein's app is most likely because he is also using PCI 3.0, which I am waiting until Ivy comes out at end of month to get mine.

WCG will be a different story however, since it is OpenCl. I am looking forward to seeing how it runs, but an also somewhat hesitant at the same time. As with single Einstein apps, unless WCG loads multiples into one task, and allows CPU to only run at beginning and end (i know knreed posted on this somewhere), the additional power may be lost, because so much time will be spent idle, waiting for CPU to finish its portion of the job. Like how the 7950 and 7970 were able to finish in what, 15-25 seconds, but WU still took a couple minutes to complete if I recall correctly. On this note, skgiven will be correct in the fact that having multiple "average" GPU's will be FAR more productive.

And of course on FP64 tasks, this thing is TERRIBLE. But that was known when I got it (only MW and Prime tasks (non sieve) use it though i think so i didn't really care).

Looking forward to some betas, whenever more NVIDIA are released, I think they only came out on first batch? At any rate, its performance/watt smokes a 580, which is ALWAYS GOOD

EDIT: Figured I'd mention this as well. Many people on different forums are VERY upset with the fact that it's a 104 board and not the 110 that many people were expecting. Some people think that a 685 or whatever will come out featuring this board, but one thing I can say, is that the design of the 680 is not intentionally crippled in the sense that they "turned off" FP64 power for lack of a better word, but there's only eight SEPERATE FP64 cores ON THE BOARD that are not counted in the total. It's not that it's crippled b/c they wanted to, it was to keep the FP32 skyhigh, which they accomplished. If a big board does come out, I personally doubt that NVIDIA would change it's ENTIRE gtx line architechture (spelling def. wrong). They have reserved these (more FP64 cores) for their expensive quaddro and tesla series would be my guess. Games don't use it, and we crunchers who buy desktop boards for crunching are a VERY VERY small minority. So, from a business perspective, IMHO I think they made the right move. Not a fan of the boost though.