|04-28-10, 05:50 PM||#1|
Join Date: Apr 2010
3D Vision with Mitsubishi DLP Problems
We have been trying to get 3D Vision glasses and emitter going with our Linux box since the new drivers came out supporting it under linux (vers. 195.36.15 ). We are following the "Stereo" X configuration documentation which is found in the README that comes with the driver.
When we add Option "Stereo" "10" - for 3D Vision stereo, we could not go into 1920x1080x60hz because for some reason, X doesn't even attempt to validate the EDID modes it reads from the DLP display. It validates these modes with no Stereo option, but ignores them otherwise. Also, there are warning messages that the EDID contradicts itself, and in fact it does (listing 1920x1080x30hz as an acceptable mode, but horiz refresh rate is 59-60 hz so maybe the EDID is wrong). At any rate, I mention this just in case someone has a similar problem. Also, when choosing the Stereo option number 10, X starts reporting that the native resolution of the display is 640x480. Another forum post mentioned that the NVIDIA driver is hard coded to use the first EDID value as the native resolution (which is the case for us). We get around all of this by disabling EDID and using ExactModeTimingsDVI; using the modeline from EDID that worked when we had no Stereo option. We are then able to put the display into the correct 1920x1080x60 resolution with Stereo enabled.
However, through all of this, we still have one huge problem. It is the error message below:
NVIDIA(0): --- Building ModePool for DFP-0 ---
NVIDIA(0): Unable to find any NVIDIA 3D Vision Stereo mode timings for
NVIDIA(0): use with the EDID of DFP-0; NVIDIA 3D Vision Stereo may
NVIDIA(0): not function properly. This might happen if no EDID is
NVIDIA(0): available for DFP-0, or if the NVIDIA 3D Vision Stereo
NVIDIA(0): mode timings were invalidated, e.g., if a display port
NVIDIA(0): connection lacks adequate bandwidth.
NVIDIA(0): Native backend mode 640x480 is not compatible with NVIDIA 3D
NVIDIA(0): Vision; NVIDIA 3D Vision may not function properly on
From what I can tell, this implies that there is some set of 3D Vision Stereo mode timings which are valid. But I cannot find a list of what these valid timings would be. Also, note that even though I have explicitly told X not to use EDID (with the Option "UseEDID" "false" directive), the 3D vision section is still trying to use 640x480 as the native resolution (I guess because the EDID is wrong on the display). If we ignore the warnings above about 3D Vision, and try to run "glxgears -stereo -fullscreen", the LED on the emitter lights up solid green, and the display does in fact show two images. However, the glasses don't flip at all (even with ForceStereoFlipping), and the image on screen is jittery (i.e., when we pause the gears from rotating, you can see the flicker from it trying to draw both images sequentially). It looks nothing like what it would under windows when it is working (or like it would running glxgears with other stereo emitters and headsets).
1. Is there any way to make the NVIDIA driver stop using EDID for the 3DVision stereo mode 10?
2. Is it complaining because the mode we set is at 60hz? Does it in fact HAVE to be 120hz input?
From our understanding, the NVIDIA driver creates the Left/Right image at the driver level, and interlaces them in a checkerboard fashion and sends this to the display at 60 frames per second. On the DLP, the display recognizes this, and pulls the image apart and (in Mitsubishi's case) uses the Smooth120 technology to show each of left/right images in succession (60hz x 2 = 120hz). We are not using a 120hz modeline because we thought this would require a dual-link DVI-HDMI cable , but the 3D Vision glasses come with a single link DVI-HDMI cable. Also, when we tried constructing a 120hz modeline, we were told by the driver that the pixel clock was over our max pixel clock (of 330). Further compounding the confusion is the fact that the NVIDIA driver spits out the following:
Connected display device(s) on Quadro FX 5600 at PCI:2:0:0:
(--) Apr 28 16:00:50 NVIDIA(0): DFP-0
(--) Apr 28 16:00:50 NVIDIA(0): DFP-0: 330.0 MHz maximum pixel clock
(--) Apr 28 16:00:50 NVIDIA(0): DFP-0: Internal Dual Link TMDS
(--) Apr 28 16:00:50 NVIDIA(0): DFP-0: Native FlatPanel Scaling is not supported
(--) Apr 28 16:00:50 NVIDIA(0): DFP-0: DFP modes are not limited to 60 Hz refresh rate
(--) Apr 28 16:00:50 NVIDIA(0): DFP-0: DFP is not internal to notebook
But the Maximum pixel clock reported from the EDID is 170 Mhz. Wouldn't this mean we need a dual-link DVI-HDMI cable? After all, HDMI supports dual link tmds.
Lastly, this all works perfectly in Windows. Same graphics card, display, set-up. So if it works in Windows, it can't be the cable that is the problem.
Sorry for the long post, but hopefully it will be helpful to others, even if we receive no answer.