Gaussian 98 A7 Processor Comparissons (1 CPU & 2 CPU)
Job = LADH Dimer (76723 Atoms) QMMM CIS/3-21G* QM on NADH residue
Gaussian Input file is here.
Route = #p CIS=(Direct, Nstates=4,conver=3)/3-21G* Charge Nosymm
%Mem = 256MB
Compiler options used:
SSE ON = PIII & AMD Athlon XP = PGF77 (3.2-4a) Portland Blas Library -Mvect=sse -r8 (-tp p6) 256 KB Cache
SSE OFF =
PIII & AMD Athlon XP = PGF77 (3.2-4a) Portland Blas Library -Mvect=256
KB Cache (-tp p6)
AMD Athlon = PGF77 (3.2-4a) Portland Blas Library -Mvect=prefetch (-tp athlon) 256KB Cache
|
System Specs Dual Intel PIII 800EB Dual AMD Athlon 1.2GHz MP AMD Athlon 1.4GHz AMD Athlon 1800XP 1.53GHz |
Timings
| Processors | Time for Optimisation (s) | Speed as Function of 1xPIII 800 (time for PIII800/time for Proc) |
| Single Processors | ||
| PIII 800 | 20955.6 | 1 |
| AMD 1.2GHz MP | 14345.7 | 1.460758276 |
| AMD 1.4GHz | 13469.3 | 1.555804682 |
| AMD 1800XP 1.53GHz SSE on | 13260.7 | 1.580278567 |
| AMD 1800XP 1.53GHz SSE off | 12205.7 | 1.716869987 |
| Dual Processors | ||
| 2 x PIII 800 | 14916.5 | 1.40486039 |
| 2 x AMD 1.2GHz MP | 8421.3 | 2.488404403 |
Conclusions:
The athlon 1800XP chip is faster than the 1.4GHz Athlon by a factor of 1.103 cf. the ratio of the clock speeds which is 1.093. Thus The new chip design is indeed faster than the old chip design but hardly by the sort of amounts AMD are touting. How they justify the 1800+ tag I don't know. Apparently this is supposed to be a comparison to the standard athlon range and not as is commonly misquoted a comparison to Intel P4's. Thus the PR department would seem to be stretching the truth a bit since a 1.8GHz standard athlon would probably be a lot quicker than the 1800+XP chip.
SSE
SSE on the PIII 800 works very well giving a speed up of around 12% for Gaussian. The Athlon XP now includes SSE extensions but they would appear to be pretty rubbish since turning them on actually slows down Gaussian.