There are two "cores", seperate mathematical engines for different protein projects in use at present, the client downloads the one it needs automatically. They are known as the tinker core and the gromacs core. With tinker core projects, the Athlon XP is undisputed king, and clockspeed is the most important factor. You will find many people with farms using XP1600s or morgan core Durons with SDRAM boards. However, with the Gromacs units, with a commandline parameter -advmethods the P4s (and P3s compared to same clockspeed Athlons) are faster because of higher clockspeeds and better SSE instruction units. However, that said, XPs are no real slouch on gromacs, P4s just edge them out.
I guesstimate that about 30% of the work is gromacs overall, tinker units being more common. There are times when you can keep the gromacs units running 24/7 but then they run out and it's back to tinker. I do not know if there will be more emphasis on gromacs or tinker in the future.
So what's performance like? I'll guesstimate that on tinker units, an XP Mhz is worth nearly 2 P4 Mhz, well a closer guess is about 1.7 I think. So to outfold an XP1600 (1400Mhz) on tinker you'd need a P4 2.4. I think on gromacs the XPs perform just about level with thier clockspeeds or slightly under their XP ratings, so an XP1900 would match a P4 1.6, the only reason P4s are particularly better really being that you can get ridiculously high clockspeeds with them. So on average, if the distribution of Gromacs units stays about the same, your main rig at default speed would probably outproduce a P4 2.2 by 40%. If the work went to being all gromacs, clock for clock I think it would be about even.
However, even if it came about that everything was all gromacs, the price/performance advantage is definitely in the XPs court. Since you could use 2 lower priced XPs in a dual to get a better overall Mhz than a single top of the line P4 for the same price in motherboards and CPUs. (P4+board=2XP+dual board)
Basically, XPs rock for folding, but we'll take anything you've got
Also, in general, clockspeed is king, but gromacs shows up unstable oc's, memory bandwidth isn't all that important, but does show a minor difference, latency seems to make a measurable difference on my systems but YMMV, so CAS2=good, so run lower mem speed with better timings. It's probably latency issues with the P4 architecture than slow it down.
regards,
Road Warrior
