ACK. It would seem that the 3.53 detonators are lying about their AGP capabilities, and are in fact, pushing textures out across the system bus and not the AGP bus. This would account for the lack of a performance change when changing AGP aperture size. This could certainly lay at the root of my performance issues in these texture-intensive games. Additionally, these test results illustrate why requests for AGP memory do go answered for games that try and use the AGP capabilities of the card anyway...the downfall is that the transactions occur across the system bus (why did I ever bother with an AGP card, NVidia?). Is this some new hidden limitation of the GeForce skulking in the closet that TNT/TNT2 users inherit as a result of using these drivers? Why do these drivers continually report a constant pool of available memory, when clearly it's anything *but* constant?
I don't know the answer to all of these questions yet. I am experimenting with Detonator drivers going back to 1.09 and OEM versions as well to see if this AGP deficiency has been with us all along. More on that later. In the interim, I'm hoping NVidia has a response or explanation for these test results. [break]Nick Trantios van nVidia heeft er inmiddels op gereageerd:[/break]There were some issues in Windows 2000 that are being fixed by Microsoft, but are not in the latest public release of Win2K. My guess is that this user is hitting one of those issues.
In addition, the state of AGP on Win2K is that it is in an early stage of development. NVIDIA is still working through all the issues in both our code and Microsoft's for handling AGP in the same way that we handle it on Win98.