From the orifinal author, this matter seems to be getting a bit old by now. original message:
simulations that require a lot of calculations, typically taking up to a week on a fast P III machine (program Paup*). By all accounts the AMD Athlon 1.4 GHz is faster than P4 1.5 GHz 512 MB without DDR RAM and maybe about equally fast as P4 1.7 GHz with an equal amount of RAMBUS RAM or at least DDR RAM. But the 1.7GHz is a lot more expensive. However a I talked to a person who claimed that on his set up a P III with 1 GHz beat Athlon in Linux in an application that required a lot of RAM and calculations (he had 750 MB).
I believe that the problems this person has been experiencing with Athlons were not general. Maybe due to hardware problems: Athlon should have a 250W-300W power supply but the vendor was selling them with a 200W power supply. This could have effected the persormance if for instance the processor heat up and cut the clock speed in half etc... He also tried to build a Beowulf cluster (44 nodes) out of Athlon and that did not work and he had to switch to P III :- ( . Maybe the Beowulf software did not support Athlons.
1.) Can anybody confirm this from their own experience?
I've got a 1.4Ghz Athlon (just in) which is about 1.5 times faster than a 1Ghz PIII (with our own software (seismic data processing).
I have confirmed from the maker of the software in question (paup*) that AMD Athlons are about equally fast or faster than P4 on this program as well. There aren't official benchmarks on this however. So Athlon seems to be a very good alternative for scientific computing. Hopefully they get the Beofulfs also working for Athlon. P4 vs. PPC: PPC speed: The bench mark results. The run represents a typical, although a bit shorter than usual run on the PAUP* software. The software has been originally made for MACs and the code has been optimized for G4 Altevac (alt). The difference btw P4 1.7 MHz (and presumably AMD 1.4 GHz) is quite clear and also Alpha 21264 performs worse. The optimazation for G4 code can be seen when comparing the port (portable eg the same code as in P4 Linux PH 7.2) run under OS X and alt (Altevac) run under OS 9.x: 11 hrs vs. 8.8 hrs. Or if the Apple native code was run under G4 800: approx. 9.5 hrs vs 8.8 hrs. P4 vs. G4 PPC runnning the same code: 11 hrs vs 5.5 hrs. An exactly 2x difference. There seems to be an almost linear relationship to the clock speed, although this does seem quite strange. The benchmark was made by the people who developed the software and optimized it for G4. G4 with lots of L1 and L2 cache is probably great for graphics applications.
P.S. This could be the reason why MAC G4 800 was left standing in the dirt by P4 1.7 GHz, even thought the code was optimized for G4 and not for P4.
Where did you read this ? The latest benchmarks in the C'T show that the G4 (esp. with code optimised for the G4) is about as fast and sometimes faster than a P4 ...(P4/1.7)
Benchmark (time in hrs.): this has not been officially published yet ...
Source: Jim Wilgenbusch