DK8N, 6800GT & 4GB mem
I just bought a new computer. It's based on the Iwill DK8N NForce3 250 motherboard. It has 2 Opteron 250s, 4GB of memory (4x Corsair 1GB Registered ECC DDR400, 2 per CPU, dual-channel, NUMA) and an XFX GeForce 6800 GT. I bought it specifically to run Linux x86_64.
So far, I've had more than a little trouble with the graphics card. I'm running Debian pure64, but I've also tried Fedora Core 3 test2. I have similar problems on both.
My first problem is that the NVidia x86_64 GPU drivers don't install on pure64 very well, but I managed to solve that by extracting the archive with "-x" and installing by hand. I think this is a known problem, but it should be fixed...
My second problem (and the reason for this post) pertains to AGP. I can't make the card run in AGP mode with 4GB of ram in the system.
The Linux AGPGART (amd64-agp) fails completely on this machine. I'm not at the machine at the moment, so I'll have to post specific logs later (if requested) but it complains that the AGP aperture is pointing to RAM, and that it can't allocate a 4096MB AGP aperture, AND that the aperture is too small, when I load the driver...
So, I tried NvAGP. This worked slightly better. The X server comes up and /proc/drivers/nvidia/agp/status shows that AGP 8x is enabled. I see my gdm login screen. If I log in, I can usually get about 1 second or so into the login process, then I get an X server hard lockup. I can move the mouse, but I can't do anything else, including switching VTs. I've had to use the magic SysRq key to reboot.
Since I've got 4GB of memory, there's no place for the PCI/AGP memory mapped IO to map to, so I loose quite a bit of my memory to the PCI/AGP addresses. I'm down to about 3.2 GB reported by the BIOS. Anyway, I started thinking that this might be the root cause of my issue with AGP and tried an experiment:
I removed 2 GB of the memory, leaving just 2GB on one of the Opterons. Then I tried the X server with NvAGP again. Everything works perfectly. ...agp/status shows AGP 8x and I can log in, and my OpenGL works as expected.
As soon as I add the other 2GB back, I can't make AGP work without locking up the system.
I don't think it's a hardware problem because Windows XP works great with the card, even with 4GB in the system.
So, First: Is this a known issue with 4GB in an NForce3 MB, using the Linux drivers? Is there a fix/work-around for it? Or, am I the only one who is having the problem?
Second, is there something I can set/change/try to see if I can make it work? Or is there a way to get the driver to output more debugging information? (There doesn't appear to be any log of a problem when it locks up.)
I'd be more than willing to do whatever it takes to track down the problem, because I spent a lot of money on this thing, and REALLY want it to work correctly. I'm perfectly comfortable compiling kernels, collecting logs, patching source code, etc. So let me know what I can do to help figure out what's happening here.
Let me know what logs/information I can post, and I will.
I would appreciate any help anybody can lend.