GF6800 Ultra with 6106 drivers -> Kernel panic

[MadEgg]

Today I upgraded my GF4 Ti4400 in a working setup with driver version 6106 to an AOpen Aeolus GeForce 6800 Ultra-DVD256.

When I booted into Linux the card did work right away, including AGP8X, but of course I wanted to benchmark to see the performance difference with the old card.
So I ran Unreal Tournament 2003's flyby-asbestos.sh in the ut2003/Benchmark directory. It zaps right trough the flyby and at one of the normally last scenes the screen locks up, nothing happens anymore.

If I press Ctrl+Alt+F1 to get to a text console, I get to see a text console but I can't type at it. A few seconds later the following text will appear:

CPU 0: Machine Check Exception: 0000000000000004
Bank 0: a200000084010400
Kernel panic: CPU Context Corrupt

I tried all kind of things to fix it. I switched from the kernel's via-agp agpgart driver to NVIDIA's built-in agp driver. I switched kernels(from 2.6.8 to 2.6.7 and different patchsets etc.
I disabled Fast Writes. I also changed the SBA setting from 1 into 0 but unfortunately that does not disable SBA, according to /proc/driver/status/agp SBA is still enabled while Fast Writes is disabled.
I also disabled the framebuffer and bootsplash. All give the exact same result at the same scene in the flyby benchmark. It did work perfectly fine with the GF4 Ti4400 though.

Does the above error message mean anything to anyone, can anyone help me to fix this?

[SuLinUX]

Looks like your CPU is erroring or on it's way out, do you overclock?

[MadEgg]

Originally Posted by SuLinUX

Looks like your CPU is erroring or on it's way out, do you overclock?

I used to before I put the new card in, I never had any problems with it. But when I got my new card I put the FSB back to normal to avoid any problems. Obviously it didn't work.

[blueworm]

What power supply are you using?
Monitor the +12v rail, Obviously sensors output is not reliable unless it is calibrated.
But this said if you are suffering a voltage drop under heavy load you will see a fluctuation.
Try running prime95 torture test to test CPU.
Also test memory with memtest86.
To put GPU under load you will find that glxgears will do fine if not try SPECVIEW perf.
BTW all these programas ARE available for linux.

[MadEgg]

Well, I've only got a 300 Watt power supply with 2 12v power cords of each max 15A. I heard that to connect it to one cord you need at least 18A on each line so I used one molex connector of both 12V power cords to power the card.

It could be that it has anything to do with that but I don't know how I can find that out...

Besides, I don't have any problems with the card except for a bit low performance(13000 score in 3dmark 2001se vs 28000 common for the card, 10000 in 3dmark 2003 vs 13000 common for the card). Hmm, if I read that again, might be the same problem then.

I'll try to get my hands on a new, stronger power supply to see if that helps.

[MadEgg]

Right, I replaced the 300 Watt PSU with a 450 Watt PSU, should be more than enough.

The situation only got worse now though. Whenever I try to run the benchmark, I get a black screen and that's all I'm going to get. I don't even get to see anything UT2k3 related anymore. :-(

[SuLinUX]

That error is pointing directy to the CPU, UT2004 is VERY CPU intensive and under high load the Kernel is picking up CPU errors.

[MadEgg]

Define 'CPU Error'... Are you saying my CPU is broken?

I tried yet another PSU, a 650 Watt PSU this time, and I still get the same problem, directly after running ./flybe-asbestos.sh I get a black screen and nothing is ran.

In Windows XP I don't have any problems with it though. 3DMark2001SE/3DMark2003 give reasonable results, UT2003 runs like a charm(although the framerates aren't exactly high on medium quality settings(40-60 fps), they stay about the same when putting everything on highest quality @ 1280x1024 so I'm quite satisfied.

How come it isn't working in Linux then?

[MadEgg]

*bump*