It's a pity then, that some (or even most...) of the apparent increase of productivity is due to the horribly inaccurate cEM measuring unit
(larger n's generally generate larger cEM results-I know this isn't correct in some cases ->crossing FFT bounds?<- but this is the general tendency)
My Athlon 3000+ shows now about double scores compared to those I got 9 months ago, when I first put this machine on the project and this is definitely not due to an SSE2-enhanced P4 client or any magic increase of the abilities of my CPU
We need a revised unit soon, or we are basically comparing fake data-the "equivalent power" estimates lately show multiGHz machines that will probably not be invented for a few more years
PS: They ain't building the cEMs anymore like they used to!
(I am a member of TeamRetro after all...)