“This January not only brought new Apple systems, but also a MAC OS X-adapted benchmark suite by the Standard Performance Evaluation Corporation (SPEC) entitled CPU2000. On the one hand, this suite allows comparisons to be made within a certain framework with the Intel competition and, on the other, it shows that Motorola and Apple were able to get more out of the new gigahertz processor than might have been expected by simply taking the pure clock frequency difference to the 866 MHz predecessor model into account.” Apple G4 Dual 1 Ghz against a single PIII at 1 Ghz. Which is faster? Heise has the SPEC benchmark results. SPEC is known to be very precise when comparing the CPUs themselves without having major interference from the rest of the system or surrounded hardware. Our Take: AFAIK, the MacOSX license specifically states that no benchmarks results of any kind are allowed to be published. Coolio. UPDATE: Read more for some commentary on the results.We see that the G4 can’t always keep up clock per clock with a Pentium-III and Heise didn’t even use gcc3 or icc for the PIII… and that’s only in integer.
In floating point, the G4 significantly lags behind the PIII, and this is an area where the P4 (and Athlon) are known for being significantly faster than the PIII.
A quick comparison, when using the better compilers for the
Athlon 1666 (2000+) : 697
P4 2200 : 790
G4 1000 : 306
PIII 667 : 310
Floating Point Results:
Athlon 1666 : 596
P4 2200 : 779
G4 : 187
PIII 667 : 222
For the people who argue that Altivec was not enabled. This is true, but it is also unfair.
The compiler they used, gcc 2.95.2, doesn’t know how to use MMX or SSE either, and barely knows how to use the PPro floating-point instructions FCOMI and FCMOVcc.
Furthermore, what most people care about is to write
high-level code and to compile it, trusting that the
compiler. Going even further, most companies will want
their engineers to spend more time optimizing an x86
version than a PPC version (writing a few core routines
in vector assembly), so it’s even more important for a PPC compiler to be able to auto-vectorize code.
On SPEC, what matters is how fast the exact same (source) code runs, and nothing else. CPU vendors have no excuse for not providing good compilers.