8800GTS SuSE 10.2 severe stability issues

[bjorneven]

Hi everyone

My system locks up completely after a couple of minutes of work, especially when it's being active (like wiggle the mouse, open some firefox tabs, closing them). By lock up, I mean complete lock up with no means of ssh, even caps lock on the keyboard wont turn on..

Some specs:
POV 8800GTS
AB9 Pro
Chipset: ICH8R, in AHCI mode (no IDE emulation)
A single Seagate SATA harddrive
Dell 2007WFP connected via DVI
2GB OCZ PC6400 Platinum RAM.
C2D 6600
2x Integrated Realtek 8168 PCIe network cards
Latest Nvidia driver from source.

The system is not overclocked, and Windows runs just fine.




So, I have tried the stability tips mentioned:
pci=nommconf idle=poll vga=normal

same problem.

I have run xorg with the "vesa" driver, and everything works stable (apart from being extremely slow). The "nv" driver doesn't work.

I have tried other linux-distros, but Ubuntu and Slackware wouldn't boot the installer. I'm aware of some compatibility issues with the ICH8R chipset, but that should only be in IDE-emulation-mode, not in AHCI-mode as that is an open specification (as per http://linux-ata.org/driver-status.html#ahci)

I have also run memtest86+ about 10 times, with no problems reported. UPDATE: I am also running the latest BIOS for my motherboard.

I'm not running Compiz, but I see that SuSE has a built-in extension of COMPOSITE; I haven't tested turning it off yet (realised that now). Let me come back to that, and if it helps I'll mention it and close this bugreport. UPDATE: It didn't help.

Note that I couldn't get Xorg up with 32bits, it's running in 24bits. It should of course be able to run in 32bits..

[netllama]

I have a few questions:
0) Are you able to setup a serial console to capture kernel messages at the time of the hang?
1) Can you test whether a kernel.org 2.6.19.x kernel makes a difference?

Thanks,
Lonni

[bjorneven]

I have tried 2.6.19 and 2.6.20-rc4, without any difference.

My machine does not have a serial-port I'm afraid


UPDATE:
I've set up remote logging (next best to a serial port I guess) over UDP, but nothing interesting in the logs, unless

Jan 25 22:53:22 couchpotato kernel: SFW2-INext-DROP-DEFLT-INV IN=eth0 OUT= MAC=mymacaddr SRC=64.233.183.104 DST=10.24.69.188 LEN=40 TOS=0x00 PREC=0x00 TTL=239 ID=15758 PROTO=TCP SPT=80 DPT=19477 WINDOW=8190 RES=0x00 ACK FIN URGP=0

is interesting (SuSE firewall).

Otherwise there is nothing reported by the kernel before the crash. That could of course be related to the fact that this is networked logging.

[bjorneven]

Update: Disabling SMP in the kernel solves the problem of lockups.

But, well, now I'm only running on one core :/ I also experienced lockup while compiling NVidia-driver from source @SMP-kernel, but I've compiled several linux-kernels without any form of lockups, only nvidia-driver compiling. I've used the Vesa-driver up until now on SMP and have not experienced any lockups.

I'm currently using 2.6.20-rc6 without SMP.

[netllama]

Have you verified that the instability is still present with an SMP 2.6.20-rc6 kernel?

Thanks,
Lonni

[bjorneven]

Yes.

Although, I thought it was fixed there for a while. Had it running for about 30 minutes, but then it locked up. Rebooted and then it locked up at once.

2.6.20 doesn't work at all with my motherboard (rc7 didn't either), I guess it's a buggy release.

[bjorneven]

Reinstalled with Debian Etch, and running kernel 2.6.20 SMP.

Works like a charm with no lock ups.

Seems odd that Suse doesn't work, maybe some kind of program installed in the distribution or the compiler they used has issues with my configuration.

[Shatrat]

Originally Posted by bjorneven

Note that I couldn't get Xorg up with 32bits, it's running in 24bits. It should of course be able to run in 32bits..

I don't run SuSE anymore and I haven't upgraded to the 8800 series (yet) so I can't comment on your main stability problem, but I thought I'd mention that what is usually called 32 bit color is just 24 bit with an 8 bit buffer or alpha channel.
So, 24 bit is the same as what is often called 32.
Just run 24 and don't lose sleep over it.
2^24 = 16777216 so with 24 bits you could have a different color on every pixel on a 19 inch LCD and still have room to spare.