In my last look at productivity benchmarks, it became obvious that using Winstone to measure performance has a few tricks. Winstone is very disk limited, and not all south bridges are created equal. In order to get around this, I ran each benchmark twice, once with a the normal South Bridge, and once using a Promise Ultra-100 PCI IDE controller. Doing this lets me isolate the I/O “fog factor” – because its the same for all players, and we can see the effect of the memory/AGP controller on system performance. But remember, the overall performance indicator is achieved by using the motherboard’s I/O controller, unless of course you are going to run out and buy an IDE controller to use. I will refer to these two graphs as Normalized (using the Promise Ultra-100) and System (using the on-board IDE)
It would appear that Business Winstone is not influence by memory latency or bandwidth at all. There is very little in it. And this really is as expected.
The tweaked boards eke out a tiny gain in performance, but not much. Again it doesn’t appear that the tweaks are doing much for performance. This is as expected. Remember that the caveat I expressed was that most (>90%) of all memory accesses are satisfied from cache.
Overall System Performance
As I mentioned previously, the I/O performance plays a significant role in Winstone scores. So the scores here are to determine how the motherboards perform as a system. In an update to my KA266 vs. KT7A article, I measured and commented upon the poor performance of the ALi drivers under Windows 200. It’s not great, but VIA does appear to be stealing a march on ALi.
To put things in perspective, I’ve included two graphs that illustrate the performance gains from using the Promise Utra-100 controller versus the on-board IDE. Note: these graphs aren’t to scale!
In both cases the VIA 686B equipped KT7A and 8K7A gain about the same performance boost, around 4 Winstone points. The ALi gains much more in the Business software scores, which we suspect to be more I/O bound. Content Creation scores gain only half that amount.
In the wash though, there isn’t enough performance difference to choose a winner.
Be the first to discuss this article!