|10-19-06, 12:54 PM||#1|
Join Date: Oct 2006
DVI hot plug issue with FX 4500 X2?
Iím working on trying to get an FX 4500 X2 working with an IBM T221 (model DG3, Iím pretty sure).
Essentially my T221 has 4 single link DVI inputs (A1, A2, B1, and B2). It returns EDID on A1 and nothing on the others. Iím able to overcome the EDID issue by forcing my own mode line to be accepted using the usual techniques in the NVIDIA Linux driver.
The real problem that Iíve found is that it seems that the T221 doesnít generate proper hot plug events when it is connected to the DVI connectors of the graphics card. Iím using the following line for both GPUs to tell the gfx card to expect the following display to be connected to the first channel of both GPUs, but it seems like the card never gets the hot plug event from the display so it doesnít bother turning on the TMDS.
Option "ConnectedMonitor" "DFP"
With all that said, is my line of thinking correct in that the NVIDIA cards with the above setting will set up the display to draw whatís requested but not turn on the TMDS or DAC until the graphics card detects a DVI hot plug event? If so, how do I turn that off?
Oddly enough, this display works perfectly on ATI FireGL X2s since I can tell the ATI driver to blindly send out the pixels and not worry about detecting what itís connected to.
I am able to set up the xorg config file to draw two tiles of 1920x2400_25. If I plug A1 into the first channel of the first GPU and A2 into the first channel of the second GPU, only A1 comes through at 1920x2400_25.
If I use a Gefen DVI Detective to clone the EDID of A1 and put it in between A2 and the first channel of the second GPU, I am able to get two tiles of 1920x2400_25 as desired. Iím also able to replicate this with Windows XP too. Even though this works, Iíd rather not use the DVI Detective if I donít have to because I want to take this further where I get all 4 inputs working at 1920x1200_41 and donít want to have to purchase 3 DVI Detectives at $70 each for each system.
If I start X with nothing connected and then plug in A1 and A2 with and without the DVI Detective, the T221 display light goes from solid amber to blinking amber (no green). This leads me to believe that the T221 isnít sending out the right hot plug event and the NVIDIA card doesnít know that it should send anything out the DVI connector.
Anyhow, it looks to me like the hot plug is the issue. If true, is there any way to leave the TMDS turned on all the time and not look for a hot plug connection?
I attached the nvidia-bug-report.log files where I used and didnít use the DVI Detective.
Iíve also heard that some have had a lot of success with the NV40 GPUs, but there were some changes to the G70 based cards causing what Iím observing. I donít have an NV40 nearby to verify this however.
|10-19-06, 01:35 PM||#2|
Join Date: Dec 2004
Re: DVI hot plug issue with FX 4500 X2?
I have a few questions & comments:
0) I'm not sure if you're aware, but the IBM T221 has quasi-programmable EDIDs. The fact that you're only getting an EDID detected on one of the four DVI inputs leads me to wonder if perhaps the unit you're using isn't programmed correctly. Can you comment on this?
1) Using the ConnectedMonitor option instructs the driver to send 'pixels' from the graphics card with the assumption that its driving whatever display device was provided as a parameter to ConnectedMonitor, regardless of what is or is not detected. If I'm understanding correctly, this is the precise behavior that you're requesting?
2) You stated that "I’m also able to replicate this with Windows XP". Can you clarify what you're replicating with WindowsXP? Is it the same problem that you're reporting in Linux with the 4500X2, or some other problem or behavior?
3) Please note that there are some known issues with the T221 panel's built-in i2c driver, where it does not have the drive strength necessary to fully pull down the IIC data bits to 0V, but instead only drops to around 1.2V (from 5V). In many (most?) recent NVIDIA graphics (including those with G71 GPUs) cards the tolerances have been reduced to the extent that the T221 will fail to work at all as a result of this i2c issue. This is very likely why you've heard reports of everything just 'working' with NV4x based graphics cards.
4) Have you tested this with the more recent 1.0-9626 driver as well?
5) I'm not entirely clear on the full scope of the problem you're reporting. Is the issue that you're not getting anything displayed for A2 even if its connected to the graphics card starting from when the system is booted, or is it just if A2 is connected at some time after X is started?
I'd also like to see a verbose bug report against 1.0-9626. You can do this by starting X with this command:
startx -- -logverbose 6
and then generating a new bug report.
|10-19-06, 03:09 PM||#3|
Join Date: Oct 2006
Re: DVI hot plug issue with FX 4500 X2?
Thanks for the fast follow up!
Regarding (0) and the other comments, I agree that the IBM T221 is kind of funky and non-standard given that the display was very bleeding edge for its time. Iím pretty certain that it is working as expected despite issues like lack of EDID. I was able to look in the manuals and the EDID settings are set to 35 which is for the Quadro4 900XGL which appears to be the latest and greatest NVIDIA card at the time of the printing of the manual. I also have it working great with the FireGL X2 cards. It works beautifully.
For (1), you are absolutely correct. I want data to always come out the DVI port and not let the card worry about whatís connected to it if anything.
For (2), I am able to get 2 stripes of 1920x2400_25 with both Linux and Windows XP when using the DVI detective in between A2 and the second GPU. If I remove the DVI Detective and connect A2 directly to the first channel of the second GPU, only A1 comes through giving me one stripe of 1920x2400_25 centered in the screen.
For (3), this could be the issue Iím observing. Are there any workarounds for this?
For (4), I havenít tried 9626 before my original post, but I did just try it and Iím attaching the log files as requested. Is this a beta driver or an official release? I was thinking that Iíd run into more problems than fixes if I introduce beta software to the mix. I also see that NVIDIA just released 8776 on the ftp site where I was using 8774. Does this have any fixes beyond the buffer overflow security issue?
For (5), with no DVI Detective, only A1 is detected and receives pixels. If I connect, A2, B1, and B2 directly to the DVI connectors on the card, they are not detected and receive no signal. If I clone A1ís EDID information with a DVI Detective and put it in between the graphics card and A2, I can get A1 and A2 to display properly instead of just A1 by itself. Also, if I completely unplug the T221, start X, then plug in A1 and A2 to the graphics card, I get nothing on A1 and A2 even with ConnectedMonitor configured properly and even with the DVI Detective. I do not have this problem with "normal" modern flat panel monitors, so I'm wondering if the graphics card doesn't detect a DVI hot plug event so it doesn't bother powering up the TMDS or DAC whereas the ATI cards can be set up to always send a signal out and never power down their TMDS or DAC and wait for DVI hot plug events to power up.
Thanks for taking the time to help me Lonni!
|Thread||Thread Starter||Forum||Replies||Last Post|
|NVIDIA Responds to Reports of Kepler V-Sync Stuttering Issue||Rieper||NVIDIA GeForce 600 Series||13||03-03-13 11:56 PM|