Maybe you can still remember the article Motherboard Comparison Epox EP-7KXA vs. Asus K7V , in which I pointed out the problems that Windows2000 has with the CPU-to-AGP bridge of the VIA Apollo KX133. Since this time we've done a whole lot of testing with Athlon-motherboards that are based on this VIA chipset, as found in our recent KX133 Motherboard Roundup and we've found out a lot more about this annoying issue.
What We Know So Far
Operating systems like e.g. Windows98 or Windows2000 recognize the different components on the motherboard of a system via the device numbers that are reported by the motherboard BIOS, something that I already pointed out in the above mentioned article. This is not necessarily too much of an issue, because many of those motherboard components work just fine, regardless if the operating system recognizes them or not. In case of VIA's CPU-to-AGP-bridge in combination with Windows2000 the story is a bit different. There simply won't be any AGP-support, if the CPU-to-AGP-bridge doesn't get recognized by Windows 2000. This situation is obviously unacceptable, since e.g. frame rates of no more than 70fps in Quake 3 Arena are the result, regardless which CPU is being used.
As you remember, VIA's KX133 reference board as well as Asus' K7V motherboard both report '8598 ' as the device number of their CPU-to-AGP bridge, while Epox' 7KXA is reporting '8391 '. Unfortunately Windows 2000 doesn't care much for '8371', so that you won't have AGP-support under Win2k with the 7KXA, unless you change the file'machine.inf' as described in the above mentioned article. This situation led me to the conclusion that Epox was doing something wrong. Our KX133 board test showed however, that the situation is very different. Epox is not at fault.
Little Tricks Vs. Insufficient Support
Believe it or not, '8391 ' is the actual correct device number of the CPU-to-AGP-bridge of VIA's Apollo KX133. '8598' is the number of the CPU-to-AGP-bridge of VIA's good old MVP3 Socket7-chipset! VIA as well as Asus were/are using a trick, which takes advantage of the fact that it doesn't matter which of VIA's CPU-to-AGP-bridges is actually being used by a system, since VIA's AGP-driver is the very same for either of them. As a matter of fact, the Apollo KX133 was NOT implemented into the Windows2000 driver support, despite VIA's claims on their own website. Instead you have got to make Windows2000 believe that the system is equipped with a MVP3 CPU-to-AGP-bridge, so that it installs and initializes VIA's AGP-driver 'viaagp.sys'.
We found in our KX133-motherboard roundup that only Asus was shrewd enough to use the little 'dirty' MVP3-8598-trick to make sure that AGP gets enabled in their K7V under Windows2000. All the other motherboard makers are reporting '8391', with the results that a normal Windows 2000 installation will leave the AGP disabled unless you are using the 'machine.inf'-trick, making Windows2000 believe that '8391' is also MVP3's CPU-to-AGP-bridge.