A few problems with 9742 Beta and 8800 GTS

[abelsson]

Hi!

I just got a new 8800 GTS, and thus I'm forced to use the beta drivers. I know these aren't released and fully bug free yet, but I hope it's OK if I report my problems anyway :-)

#1) No 1600x1200 for me.

I have a dual monitor setup, one LCD at 1600x1200 and one LCD at 1280x1024. Both connected through DVI. My main problem is that the 1600x1200 mode is rejected for the 1600x1200 monitor, and instead I get 1280x1024 on both monitors.

I'm no expert at reading EDID debug data, but I believe my monitor is reporting 1600x1200@60Hz as a valid mode and the max 162MHz pixel clock sounds right.

I had no problems with this setup on a 7600GT and the 8774 nvidia drivers Ubuntu Edgy has packaged up.

I also tried the following mode validation overrides: NoMaxPClkCheck, NoEdidMaxPClkCheck, NoVertRefreshCheck, but the 1600x1200 mode is still rejected.

I'm not sure why this mode is rejected, and I can't really figure it out from the log file. Help in interpreting it would be appreciated!

# 2) Restarting the X server or switching to a VT causes system crash.

After running an X-server, if I exit it or switch to a VT the system becomes unresponsive and nothing is shown on my monitors. I had to SysRq-SUB my way out of it.

# 3) Installation fails due to Ubuntu using dash.

the "if [ "$RET" == "0" ];" line in conftest.sh isn't understood by my shell. Removing one of the '=':s seems to fix it for dash.

I've attached a nvidia-bug-report.log for my mode problem.

[netllama]

Problem #3 is a known bug which will be resolved in a future driver release.

I have a few questions:
0) If you disable the console frambuffer graphics, does that help with problem #2?
1) If you boot with the irq=poll kernel parameter does that help with problem #2?
2) Does the instability persist if the vmware modules are not loaded?
3) To determine why the 1600x1200 mode is failing, I'll need you to start X with the following command:
startx -- -logverbose 6

and then generate a new bug report.

I should also note that the glx module failed to load, which suggests that the driver wasn't properly installed.

Thanks,
Lonni

[abelsson]

Much thanks for your speedy reply and excellent support!

One or both of irq=poll or turning of the framebuffer seems to have helped with the instability as I can now merrily restart X and switch to VT:s, I'll report back which one fixed the problem. The vmware modules seems to not cause any problems.

Ah, one log level too little. I ran the previous report at 5 :-)

I've attached a new bug-report. From my amateurish reading of the log file, it seems to first think 1600x1200 is valid but then only to reject it later under the "Native backend timings for Acer AL2021" section. Is the driver confusing the Acer AL2021 monitor's max resolution with the Samsung?

[netllama]

Try using the NoDFPNativeResolutionCheck parameter to the ModeValidation option. If that works, then this is a known bug.

Thanks,
Lonni

[abelsson]

Unfortunately it doesn't seem to have fixed the problem. But that option definitely changed the behavior of something.

The driver now claims 1600x1200 is a valid mode, but my monitor still reports it is running 1280x1024 and the text is very blurry, much more than before. It behaves as if it is downscaling 1600x1200 to 1280x1024.

The nvidia-auto-select mode for that monitor is 1280x1024 @ 59.9 Hz also, for some reason.

Thanks again.

[netllama]

According to the X log, you requested nvidia-auto-select which is 1280x1024. Did you request 1600x1200 or nvidia-auto-select? Which mode are you requesting in xorg.conf?

[abelsson]

I'm not quite sure how to check that. 1600x1200 is the first of my requested modes and I haven't done anything to explicitly select nvidia-auto-select. Except for the ModeValidation line now containing NoDFPNativeResolutionCheck, my current xorg.conf is exactly the same as the one attached in the last bug report log. I also get the same behavior with a plain nvidia-xconfig generated file, if I add the same NoDFPNativeResolutionCheck option.

The X-server definitely seems to think it is running at 1600x1200. A screenshot of my desktop is 1600 by 1200 pixels and xdpyinfo says:

screen #0:
dimensions: 1600x1200 pixels (411x311 millimeters)
resolution: 99x98 dots per inch

However, my monitor is displaying a downscaled version of that at 1280x1024. Have you encountered any kind of behavior like this before?

Thanks.

[abelsson]

Ok, if I uncheck "Force Full GPU Scaling" in nvidia-settings I get a native 1600x1200 display, which fixes the symptoms for me.

But it still displays "Native resolution: 1280x1024" for my 1600x1200 DFP.