AMD's ratings are BS, which is why a skt 754 3400 NC beats a 3500. That chip should have been called 3600 by all rights.
Well alright. since CPU speeds theorechtically scale linearly with frequency in ideal situations it's optimal to use a multiplier when comparing disimilar CPU's, disimilar speeds, disimilar feature sets etc.
Because if we were to use absolute values(like 400mhz, say for example, we know by looking at the benchmarks a Skt 754 A64 Newcastle at 1800 Mhz = a Barton at 2200 mhz diff = 400mhz which are equal performing in the global sense) we run into problems trying to apply it at differnet speeds since that absolute value does'nt scale. Natually at 2500mhz A64 you would need much more than 400 more Mhz out of the barton to equal this A64. Natually adding features like 1 mb lvl 2 or dual channel skews results further. Making absolutes a futile excersize.
Anyway just look here at the aggregates (last graph):
****Same bus speed and same mem timmings must be preserved when making comparisons cross platform.
http://hardware.fr/art/imprimer/496/
http://hardware.fr/art/imprimer/531/
You can do the interpolation But heres how it breaks down.
choose a base CPU. I'll choose Barton (B), in this case the 3200, and compare it to any other CPU finding the multiplier.
754 NC A64 x 1.25 = B
754 CH A64 x 1.30 = B
939 NC A64 x 1.30 = B
939 CH A64 x 1.40 = B
1.25 x B = P4C
1.27 x B = P4E
1.20 x B = P4EE
Want to know PR of a 2600 mhz 939 NC A64? = 2600 x 1.30 = 3380Mhz Barton or 3380 x 1.25 = 4225 Mhz P4C.
How fast is a 3200 Athlon XP compared to a P4C? 2200 x 1.25 = 2750 Mhz P4C
How fast is your OC? 2700 x 1.30 = 3510 barton x 1.25 = 4387 Mhz P4C
I think you'll find using these multipliers these results are pretty close when comparing all real world test from superpi to gaming to encoding...
rule of thumb is
A64 speed x 1.5 = P4C speed
Add 5% for Dual Channel
Add 5% for 1 mb lvl 2
Subtract 2% for skt 940 due to ECC/Reg
AXP-speed x 1.25 =P4C speed
