More Testing
While the synthetic benchmarks prove interesting, they don't always provide an accurate showing of "real" system performance. Neither does Sandra really stress a system, as it can often be run on highly overclocked components where no other benchmarking program can. With this in mind, I turned to the old favourite Quake 3 Arena, set at the default 640x480 32bit, with no sound, to test raw memory bandwidth. Again, the Radeon 9700Pro was running at 325/310 default.
What we are noticing here is that SiS655 is simply performing better the more bandwidth you give it, whether it be in Front Side Bus, Memory Bus, or ideally, a combination of both. Of course, the last two results cannot be directly compared as they are running at different CPU speeds, in this case 2.8GHz and 3GHz. It's very nice to see that going from bog standard DDR266 ram to this EL-DDR that can run DDR446 with 100% stability provides over 25 frames per second more in Quake 3 Arena at these low settings. Notice also that if you are leaving the timings detected by SPD, you won't be getting the best out of your new memory, as the jump from SPD to 2-4-3 timings show.
3DMark2001SE Build 330 was used next to further test stability at high memory speeds, as in the past I have seen this particular benchmark fail (dropping back to desktop is an example) while Quake 3 ran perfectly for hours. I ran 3DMark in a continual loop for over an hour at the highest settings possible to determine the maximum speed of this ram, on this particular motherboard (which was DDR470 at default voltage, 2.5-8-4-4).
Ignoring the higher clock speeds of the last two results, we can see an almost 1000 point increase going from plain jain DDR266 to high spec DDR446, proving that the P4 architecture can still lap up all that memory bandwidth, regardless of the fact that Dual Channel DDR266 provides a theoretical 4.2Gb/s for the P4's 4.2Gb/s System Bus.
|
