... you can now run multiple clients per card, which appears to increase their output (with the .7 beta client) to near NVidia levels! See http://forums.extremeoverclocking.com/showthread.php?t=360415 Don't be shy now! Looks like there's no reason NOT to fold peeps....
I don't think anybody every came up with a definite reason. Personally, having worked with ATI/AMD for many years, I suspect its their very basic dev rel program at fault, whereas Nvidia gives loads of support to Stanford.
The fact that running 2 clients simultaneously improves the performance (ppd wise) implies that it's a software driven constraint, so that certainly lends weight to your experience. The article gives me a bit more faith that one day someone's going to crack this and open it up to all hardware, not just NVidia's
But that's what makes me so crazy though - isn't this open source? Aren't there like hundreds of thousands of people running this and millions of AMD fans out there? Yet other open source projects get tons of free man-hours thrown into them!
No, while folding is distributed, the code itself isn't open source as (I'd imagine) Stanford needs to keep a very watchful eye on the accuracy of the results, something that wouldn't be possible if 3D party coders got involved. I'd agree with your frustration though - it's something I've communicated to AMD many times over the years, but as you know from also working with them - AMD is always strapped for resources.
Ok, but this works on the 5xx0 and 6xx0 series, with their VLIW. What about the 7970 and it's design similar to Nvidia's?