Debunking the Debunkers

Brian already commented aptly on the recent "debunking" by some random dude with a web site (obviously a key indication of expertise) of Apple's G5 benchmarks.

Rather than go into an in-depth fisking of said article, I'd just like to take a moment to destroy the already marginal credibility of its author. Because, hey, I'm like that.

First issue: The guy claims the intel side of the benchmark wasn't fair, as SSE2 wasn't used within GCC. However, had the author known anything about GCC, he would probably know that -mfpmath=sse does indeed enable SSE2.

Second issue: To the same end of the first issue, the author claims that hyperthreading was turned off on the intel box. Indeed, in this he was correct, hyperthreading was turned off for the Apple test.

It may interest some to know that Dell also disables hyperthreading for its benchmarks. Apparently, turning hyperthreading off yields faster processor benchmarks. Interesting feature.

Want some more proof? Go study the SPEC benchmarks of the Power4 processor, which the G5 is based on. With integer math, it's slightly slower than the fastest intel chip. (Not surprising, as the G5 has 2/3 the clock speed.) With floating point, it's a good deal faster. The relative trend between the two chips is reasonable evidence that if the numbers were slightly cooked, it certainly wasn't to the degree implied by the author of the previously mentioned "debunking".

Anyway, I'm getting one, and when I do, I'm going to toss Logic on it, and hook it up to an Edirol PCR-30, plug in my Gibson Epiphone, and have a grand 'ol time making music in half the time it used to take.

posted by Mr. Lion | 06/24/03 @ 19:20 | comments (0)