I have deliberately left comment on some of the Winstone results until here. If you were looking closely, you would have seen that the KT7A scored well, better even than the 8K7A+ in places. This intrigues me. Is the KT7A actually faster? Are the Winstone tests more sensitive to latency than bandwidth. I suspect they aren’t because otherwise the KA266 would have scored much worse. What I really suspect is the problem is the disk intensive nature of the benchmark. As hard disks are on the order of 500 to 1000 times slower than memory, any slight hiccup in disk access is going to magnify itself in the final score. In other words, different runs will yield different results. With that in mind, using Winstone results that vary by only a couple of points to prove Product A is faster than Product B is ludicrous. At best you can say that they came out equal.
But where is the point at which a Winstone result becomes meaningful? Unfortunately, that is for my next article. This article has thrown up the anomaly, and now it needs some investigation.
This also raises the spectre of another benchmarking nightmare. I think that I have shown that BIOS (and more particularly, memory and PCI bus latency settings) can have a much greater impact on benchmark results than some people would have you believe. Take this a step further. Manufacturer X sends out review samples of its latest and greatest. Of course it is a “preview” board, so expect a couple of crashes. But this product will have hand picked components and optimised settings. Engineers more talented than I would have spent many man-hours tweaking this beast to within an inch of its life – all in the name of performance. Is it any wonder that it performs well in the tests that Company X knows are going to be run on it. If I was willing to accept a couple of occasional crashes in the KT7A I used for this review, its numbers would be even more impressive. All I can say is wait until you see how the production parts perform before making a purchasing decision. If you can’t wait that long, buy what you can now with confidence that the production parts based on mysterious component X won’t be much (if any) superior to current products. This really is benchmarketing rather than benchmarking.
What can be concluded about the relative BIOS settings though? If you examine the different settings that I changed, you will see that the Abit has the most settings to change, and the Iwill the least. From that, it might be reasonable to assume that the KT7A has the most to gain, and certainly when compared to the default settings, the Abit makes up a lot of ground. From this it might be reasonable to assume, that for the applications tested, latencies in the chipset settings plays a bigger role in performance than whether the motherboard is using SDRAM or DDR. This is borne out by the fact that none of the applications tested showed much benefit from the increased bandwidth. If and when I can run some multimedia tests or some other tests that depend on lots of access to large chunks of main memory, then we may see a difference.
Be the first to discuss this article!