First post, by Frunzl
Hi,
I have a crazy problem and would appreciate every tiny bit of help. It is really strange, so please bear with me.
I recently got my hands on a cheap Gainward 6800 Ultra PCIe (256 MB DDR3), so I thought this would be an excellent opportunity for my first Win98 build on PCIe. The mainboard I am using is an Asus P5Q Pro with a Core2Duo 8600.
Windows installation went uneventfully and I was able to install the modded Intel chipset (ICH10) drivers. At this point, I still had an exclamation mark for the SATA controller (Marvell 88SE6111) and write speeds were not ideal (6 MB/s max).
So, after trying for days to get the onboard SATA controller to work properly, I decided to go with a separate PCIe controller, which was recognized and installed fine immediately (now 200 MB/s write speed and AHCI mode thanks to R. Löw).
This was all before I attempted to install any graphics drivers. So I force-installed the AGP driver of version 77.72 (AFAIK there is no dedicated driver for the 6800 Ultra PCIe version, PLEASE correct me if I am wrong) and the system booted fine and ran 3DMark 99, 2000, 2001SE without any issues.
However.... my benchmark results are nowhere where they should be. For 2001SE, I am getting 14000 points, where I should be getting at least 25K, probably much more...
I then thought, maybe the PCIe SATA controller is interfering with the PCIe bus or the GPU itself somehow and removed it (leaving me with the slow onboard SATA speeds).
But when I remove the PCIe SATA controller, the GPU always goes to a black screen upon bootup. No amount of trying has helped, including around 5 Nvidia driver versions, forcing the 6800GS PCIe driver etc.
It is always the same, after (re-)installing the graphics driver, the machine boots into a black screen and the GPU fan slows down (the fan usually only slows down if the driver was correctly installed, but again: the screen remains pitch black)....
By the way, the board has two PCIe 16x and two PCIe 1x slots. The GPU only "works" (at reduced speed, as described) if the second 16x slot is occupied, but not if it is one of the 1x slots. There must be a connection.
I also considered that the low 3DMark results might be due to a CPU bottleneck, and indeed the scores remain identical all the way from 640x480 to 1600x1200. But, all separate CPU-benchmarks I ran indicate that the CPU runs totally fine, exactly how I would expect. So it really seems to be a bus limitation? I have no clue...
Again, any help is appreciated!