I thought I'd better revive this thread before I have to go to page 2 god forbid!
I think we have settled on the format for the competition so it is down to the fine details.
I have been looking at the techreport page on the F@H benchmarks and they appear to be pretty good on the per WU type basis. However once it gets to overall performance I don't agree with the numbers. According to the last two graphs AMD Athlons are better than I have observed and overall a Pentium XE 965 is apparently the second-fastest CPU after the QX6700.
Now I know the numbers come from the individual benchmarks of the different types of WU but the software they use doesn't really take into account what WUs are actually available. If the best suited WUs were available then great, but as it is it would be unfair to judge say an AMD 64s performance on these numbers when during the competition there is a good chance they will not see a single tinker or amber. Briefly looking at the current project summary at Stanford, there are no Ambers and just a handful of Tinkers.
This is why I prefer the CPC numbers as they are a better reflection of the performance you can expect to achieve at the moment and it is a more exhaustive list of CPUs.
I am struggling to find a way to use the best parts of both sets of data. I don't want to dismiss the TR data as it should hold true always, whereas the CPC data is based on two projects currently available and may well change.
The only thing I can think of at the moment is rather than average the PPDs for all the WU types tested is to work out what proportion of WU types are available at the moment and maybe increase or decrease the value of each type depending on the probability of getting one...
So for instance at the moment there is no chance of an Amber so I can remove its PPD value from the average. Bad news for AMDs but you aint gonna get one anyway. There is chance of a tinker, so lets say 30% probability.. so we reduce the value of Tinkers...
It is a bit fiddly and it is just the start of an idea but I cant think of anything else - Im not a statistician
SiriusB