PCMark 2005 produces dubious results
Thrax
🐌Austin, TX Icrontian
Futuremark, the famed creator of the popular PCMark and 3DMark suites, has again given cause to question the validity of their benchmarks. Recent findings by ArsTechnica shows that a tweak to the CPUID presented to PCMark 2005 drastically changes the outcome of the memory bandwidth benchmark.
While modern chips are similar in their basic design, they are also the sum of their unique elements. These enhancements can only be leveraged thanks to the presence of the CPUID which applications and operating systems can recognize. [figure]
[/figure]
While most CPUID codes can not be changed, the new VIA Nano is free for modification. A recent review at ArsTechnica which pits the Nano against the Intel Atom allowed them to validate an old rumor that PCMark gives undue preference to certain CPUIDs.
Ars discovered that changing the VIA's CPUID to that of Intel's has increased its memory performance by over 40%. This is not evidence of fraud, but is nevertheless a damning result. The most benign conclusion that can be drawn is that Futuremark's ability to optimize code for CPUs is borderline incompetent. On the conspiratorial end of the scale, it sends a chilling reminder that the accuracy of Futuremark applications has been in question before.
While modern chips are similar in their basic design, they are also the sum of their unique elements. These enhancements can only be leveraged thanks to the presence of the CPUID which applications and operating systems can recognize. [figure]
[/figure]While most CPUID codes can not be changed, the new VIA Nano is free for modification. A recent review at ArsTechnica which pits the Nano against the Intel Atom allowed them to validate an old rumor that PCMark gives undue preference to certain CPUIDs.
Ars discovered that changing the VIA's CPUID to that of Intel's has increased its memory performance by over 40%. This is not evidence of fraud, but is nevertheless a damning result. The most benign conclusion that can be drawn is that Futuremark's ability to optimize code for CPUs is borderline incompetent. On the conspiratorial end of the scale, it sends a chilling reminder that the accuracy of Futuremark applications has been in question before.
0
Comments