edid vs. modeline
 

Hello,

in the process of figuring out how to deal with unhelpful EDID information, I came across the following which is a mystery to me. An IBM DFP connected via DVI-D to a 7900GS displays a nice 1280x1024 native resolution image when I set up xorg.conf to use EDID provided information. But when I force the driver (110.14.09) to use a modeline and not edid modes via options, the DFP refuses to show a nice display. I constructed the modeline by letting xvidtune show me the timings when the display works. And xvidtune shows me the same exact timings when the display does not work, and EDID is disabled.

Is there other information except for the timings used to drive the display (except for the content) ? Or does somehow the nvidia driver report timings to X different from what it actually finally uses ?

Here is another hint. When the display works, xvidtune reports a modeline with a 76Hz refresh rate. But the on screen information of the DFP itself says it is displaying at 59.9Hz. When the display does not work, xvidtune still reports the same modeline but now the on screen info of the DFP says it is displaying 640x480 at 59.6Hz.

The underlying reason is that on another display (3dp) I get bad edid information. This other display works when I first start up X with the IBM DFP connected, get a good display and then switch cables. So I would like to preserve that state when I start up X with the other display directly connected. The other display (3dp) does not work when I just use its EDID modes.

Did anybody come across a similar situation ?

Thanks, Andreas

Re: edid vs. modeline
 

Searching for "Modeline" in the forum gave me a bit of a history of the growing influence of EDID in the driver, and why there are all these ModeValidation options.

After reading all this my suspicion grows that perhaps when EDID is enabled the driver does pretend to use one set of timings and provides that information to X (and xvidtune) but in fact uses another set of timings to actually produce an output signal. It is hard to imagine but is it possible ?

Back in the office, I will produce nice bug reports.

Andreas

Re: edid vs. modeline
 

Here are two nvidia bug reports. nvidia-bug-report.log.edid was generated while the display on the DFP was working. nvidia-bug-report.log.noedid was generated while the display on the DFP was corrupted. In both cases running

xvidtune -display localhost:0 -show

gives identical modelines. I would like to reproduce the working output but with another device (a scaling and blending unit, 3dp) connected.

Re: edid vs. modeline
 

The NVIDIA driver will always default to using the mode timings provided in the EDID of the display device.

Re: edid vs. modeline
 

Yes, but one can disable that "feature" by the "noedidmodes" and "exactmodetimingsdvi" option, no ?

This what I tried to do. Andreas

Re: edid vs. modeline
 

Sure you can disable use of the EDID via different mechanisms. However, I'm not sure that I understand why you're trying to do that if using the EDID is producing a usable X session.

Re: edid vs. modeline
 

I would like to do this not with this DFP but with another one (see above), let's call it 3dp. The reason is that the 3dp has unusable EDID data but it works when I switch cables after starting X with the DFP connected. So I would like to reproduce the working signal, ignoring the 3dp edid data.

Does this make it clearer ?

Andreas

Re: edid vs. modeline
 

Probably the easiest thing would be to capture the EDID data from the working display using nvidia-settings and then tell the driver to use that for the non-working monitor using the CustomEDID option.

Re: edid vs. modeline
 

Yes, I have thought about that but it seemed like a hack.

So, I captured the edid.bin from the working DFP and use it with the CustomEDID option on the 3dp. It works, as expected, and gives me a 1280x1024 at 60Hz display.

How would I now go about adding on a 1280x1024 at 105Hz display, or any other mode which the DFP does not support ? The 3dp (and the projectors behind it) can show that using another image backend.

The first step may be to replicate the working mode (1280x1024 at 60Hz) but without edid, in a custom modeline. How would I do that ? Use xvidtune, or the edid timing information as reported in the Xorg.0.log ? That did not seem to work, eg. resulted in a refusal to accept input or to provide output (black screen).

Re: edid vs. modeline
 

RTFM. sorry i couldn't resist.

http://us.download.nvidia.com/XFree8...ppendix-b.html

for opensuse 'info xmode'

check for the equivalent of xmode in your distro, or just use the modeline above.

enjoy
/dev0

Re: edid vs. modeline
 

You'll need to use the mode timings of the DFP's native backend mode. You can find these in Xorg.0.log if you start X with the -logverbose 6 option.

Re: edid vs. modeline
 

Thank you very much, be assured by now I am intimately familiar with the readme.

I tried your modeline with the useedidfreq option, and it appears to work at first. But what happens is that the driver falls back to the edid 1280x1024 mode as "backend", so the modeline is not actually used. See netllamas note above. Only when I specify the exactmodetimingsdvi option, the modeline is actually used. But then I get no signal delivered, or the signal is refused, not sure which.

This happens even when I use a modeline which should be exactly equivalent to the 1280x1024 edid mode, as reported in the Xorg.0.log. This is still a mystery to me. Actually, sofar, I was not able to find any custom modeline which produced a result. That is why I asked for input.

I think I will try an older driver next. Could that be promising ? It basically means I am out of ideas.

Andreas