10. Discussion Of 64 And 32 Bit Performance Test v6 Results On Windows XP Pro X64

Look at the last seven findings in the table. These are composite scores that summarize performance for each group of tests. The greatest gains for the native 64 bit version of PTv6 over the 32 bit version running in WOW64 emulation are in CPU (18 percent) and 2D graphics (14 percent).

The PassMark Rating summarizes scores on the 25 non-composite scores; it's an overall score for all of the tests. The 64 bit version of PTv6 does about 11 percent better than the 32 bit version running in emulation mode. Looking at specifics, some of the individual CPU scores show large improvements between PTv6 64 bit and PTv6 32 bit, explaining the high score on overall CPU performance. The composite 2D graphics score is due to PTv6 64 bit's significantly better performance on the GUI benchmark. I currently have no explanation for this difference.

What does all this tell us other than that 64 bit processors don't produce twice the performance of 32 bit processors, at least using PTv6 in a notebook configured like our MS-1029? First, it's reasonable to expect a CPU intensive 64 bit native application to do better on a 64 bit operating system than a 32 bit application that can't take full advantage of the increased number of registers, wider data paths and internal cache registers of a 64 bit CPU.

Where the CPU tests improved little between the 64 and 32 bit versions of PTv6, the issue may be the very small amount of RAM (512 MB) available for data in the MS-1029. The folks at PassMark suggest that this could also account for the very small differences between 64 and 32 bit 3D graphics performance. 3D graphics applications involve lots of data and calculations. More memory is clearly better, especially for data storage. This indicates that gamers interested in the best performance from their 64 bit systems should shoot for the most RAM their motherboards can hold.

More memory might also have given the 64 bit version of PTv6 a chance to score better in the memory tests. The tests hardly get going before the meager 512 MB of RAM is filled.

It might interest you to know that there were no significant differences in 2D or 3D graphics tests between runs of 64 or32 bit PTv6 at display resolutions of 1280 x 800 (16 or 32 bit color) or 1024 x 768 (16 or 32 bit color). The MS-1029's native display resolution is 1280 x 800. As with the PTv6 3D benchmarks, PassMark says this can happen when a test benefits from lots of RAM and the test system has little of it. This can also happen when a graphics GPU driver is recompiled using a 64 bit compiler but hasn't been optimized for the 64 bit environment. I have been unable to reach ATI for input on its X700 64 bit driver. I will update this paragraph when and I have input from the GPU company.