PCMark Memory Benchmark Favors GenuineIntel over AuthenticAMD


By now, everyone's aware that Intel has the fastest chips on the market, and with Nehalem getting closer to release, the chip maker's position doesn't look to change anytime soon. But what you don't know is that Intel also has the faster name. Confused? You're not the only one.

Before clarifying, let's first look at how manufacturers label their processors. Each chip contains a processor-specific character string detailing the manufacturer, make, model, and available features. The two common ones you're probably familiar with include GenuineIntel and AuthenticAMD, neither of which can be changed. That's not the case with VIA's Nano processor (CentaurHauls) and it's here where things get interesting.

Looking to expose any irregularities in PCMark05's benchmark results, ArsTechnica ran the suite on a VIA Nano processor and then re-ran the benchmark two more times after changing the chip's CPUID to GenuineIntel and AuthenticAMD. To prevent unbiased results, a point Futuremark touts heavily with its benchmarking program, altering this string shouldn't have any effect on performance. And for the most part it didn't, but ArsTechnica did discover a troubling discrepancy in PCMark05's memory results.

By changing the CPUID from CentaurHauls to AuthenticAMD, the news outlet noted a 10 percent jump in memory performance. Even more surprising, changing the string to GenuineIntel saw memory performance jump up by a staggering 47.4 percent. To eliminate the possibility of user error, ArsTechnica claims to have "benchmarked each CPUID multiple times across multiple reboots on completely clean Windows XP installations."

Before crying conspiracy, the news outlet points out a simple explanation as a likely cause for the irregular benchmarks, and that's sloppy coding. Based on the CPUID PCMark05 reads, it could be disabling various optimizations on the assumption that they're not supported, even if they really are. Even if this is true and assuming ArsTechnica's results can be verified, Futuremark will still have plenty of egg on its face, and it might take more than a software patch to clean it off.

Do you have a different theory as to what's going on?

Image Credit: Futuremark

Around the web