Created attachment 120545 [details] first crash: Xorg.0.log As soon as I set 3840x2160 for the external monitor next to my integrated display Xorg will crash soooner or later. This reminds me of Bug 70510 which is about much elder hardware. Nontheless I would suspect this to be a memory allocation issue as well.
Created attachment 120546 [details] first crash: journal
Created attachment 120547 [details] second crash: Xorg.0.log
Created attachment 120548 [details] second crash: journal
It is really a pity that my NVIDIA Corporation G96M [GeForce 9600M GT] has been crossed out of the list of supported cards since an update to kernel 4.3.3-2. It has worked well with the nouveau driver except for this bug (which may more likely be related to memory allocation rather than just this card) and an s2ram issue (https://bugzilla.kernel.org/show_bug.cgi?id=108561).
(In reply to Elmar Stellnberger from comment #4) > It is really a pity that my NVIDIA Corporation G96M [GeForce 9600M GT] has > been crossed out of the list of supported cards since an update to kernel > 4.3.3-2. what do you mean?
When I try to start an X server with the nouveau driver it says no screens found and outputs a list of supported graphics cards; (issue has appeared with the update mentioned before).
(In reply to Elmar Stellnberger from comment #4) > It is really a pity that my NVIDIA Corporation G96M [GeForce 9600M GT] has > been crossed out of the list of supported cards since an update to kernel > 4.3.3-2. It has worked well with the nouveau driver except for this bug > (which may more likely be related to memory allocation rather than just this > card) and an s2ram issue > (https://bugzilla.kernel.org/show_bug.cgi?id=108561). I assume you're just saying this for sensational value? There's no official supported list, but in general all Riva TNT and later hardware ought to work to one degree or another. I had a long comment about potential things you might do, but given the troll-y nature of your comment I suspect that's a losing proposition. So all I'll say is "update & pray". Good luck!
(In reply to Elmar Stellnberger from comment #6) > When I try to start an X server with the nouveau driver it says no screens > found and outputs a list of supported graphics cards; (issue has appeared > with the update mentioned before). yeah well, in that case the xorg log might help
Created attachment 120877 [details] waitcard0 - script used from now on to start the X server Hi; at first I thought that there was something screwed up with my xorg.conf; however as I wanted to post an Xorg.0.log I found that the following script (waitcard0) will help. Likely this is just a configuration issue which should be reported to my distributor. Many thanks for your immediate offer to help me, Karol and Ilia! It is very appreciated.
Good news: I can now set various 3840x2160 modes without making nouveau halt or crash (It may still hang after a heavy reconfiguration session though). The only drawback about it is that my displays (AOC u2868pqu, Samsung flat-tv) will stay dark (3840x2160 non-interlaced) or in a fact show a 2560x1600 mode when I try to set a 3840x2160 interlaced mode (the monitor does not claim to support interlaced modes by its EDID though). Tested with various modelines as well as the modeline suggested by my xorg.conf. Would anyone have a look whether it should be possible to achieve 3840x2160 with this hardware configuration (Xorg.0.log is available; decoded edid will come shortly.)?
Created attachment 120878 [details] AOC u2868pqu: decoded EDID
I highly doubt your card could output 4K (maybe on perflvl 0f, but for reclocking on that card you need kernel 4.4), and even if it did, I don't think it supports HDMI 1.4 or DisplayPort 1.3(?) which are needed to support 4K at, at least, 30Hz.
Just did a small search on NVIDIA's website: only Kepler+ cards support 4K, with the notice for Kepler cards at least, that > 3840x2160 at 30Hz or 4096x2160 at 24Hz supported over HDMI. 4096x2160 (including 3840x2160) at 60Hz supported over Displayport. Support for 4k tiled MST displays requires 326.19 driver or later. Tesla and Fermi cards are advertised of having a "Maximum Digital Resolution" of 2560x1600, so no, 3840x2160 won't work on your G96M, sorry.
Now that tells us what is officially supported. I have just yielded 3840x2160@30 with an Nvidia G96GLM [Quadro FX 770M] (Windows 7) which is a very similar but better card than the Nvidia G96M [Geforce 9600M GT] as mentioned in this report. This even though my Celsius H270 just has a DVI output; I have converted it into HDMI with an adapter and plugged the HDMI2.0 cable into the HDMI1.4 port of my u2868pqu monitor. The Amilo Xi3650 notebook as shipped with the G96M card [ Geforce 9600M GT] on the other hand would have a native HDMI output (can`t tell on whether it is HDMI1.4 or sth else though). Pierre, may I come back to you after I have compiled kernel 4.4 and as far as recommended a fitting Xorg? If there was a chance to get the card running under 3840x2160@30 with perflvl 0f I would really like to give that a try. Even much more important I had some painful problems with s2ram and the u2868pqu which are device driver related (kernel bug 108561). Perhaps you can help me with this issue as well; have my full endeavor if you think that you can do something about it.
I can't quite tell what you're trying to do, but here are a few pieces of information that you may find useful: HDMI cables can only carry a single TMDS link. (Not enough pins for more.) DVI cables can carry one or two TMDS links, each maxing at 165mhz (for a total of 330mhz for dual-link DVI). The maximum TMDS link speed nouveau will allow on HDMI connectors is 165MHz. With kernel 4.5 (or anything that includes https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=1a0c96c075bb4517d4ce4fb6750ee0a3cf38714c and a few of the surrounding commits), you will be able to override this by setting nouveau.hdmimhz=anything-you-want. We have no idea where to get the limits from, so we just set it to 165mhz for tesla and earlier 225mhz for fermi 297mhz for kepler+ We'll have to figure something out for HDMI 2.0 eventually. But again, this is all with the above commits; with kernel 4.4 it's just 165mhz for HDMI. Assuming your connector and cables support it, dual-link DVI will give you the highest maximum frequency.
Great! With nouveau.hdmimhz=225 I can yield such modes like 3840x2160@23/24Hz whereby 23Hz is stable and 24Hz sometimes starts to flurr. For those who prefer a higher freuquency 2560x1600@47 should be a tip. 25MHz for 2840x2160 will make my display stay black and 48Hz for 2560x1600 results in considerable pixel flurring. I have also tried it with 3840x2160@48i interlaced. However with all 3840x2160 interlaced modes the mouse pointer disappears. I can still move the mouse highlighting the icons where the mouse moves over but the mouse pointer will be no more visible as soon as I select an interlaced mode. I don`t know whether a fix would be possible for the interlacing mouse pointer issue but that would even give a better 4K experience on elder cards as modes with 24/25/30Hz yield a somewhat deferred moving of animated objects like the mouse pointer. Nonetheless the joy of running a full 4K display with my Nvidia G96M [GeForce 9600M GT] is a thriving experience. To make this report complete nouveau.hdmimhz=297 will let my HDMI output simply stay black. The support of the hdmimhz parameter is an amazing feature that no proprietary driver will give; neat work!
By the way does anyone know whether there is a similar feature for the radeon driver? I have an XFX Radeon R5 230 card which was advertised as 4K ready. However I did never get any of the 4K modes to work.
(In reply to Elmar Stellnberger from comment #16) > let my HDMI output simply stay black. The support of the hdmimhz parameter > is an amazing feature that no proprietary driver will give; neat work! From the NVIDIA blob readme file: """ Option "ModeValidation" "string" "NoMaxPClkCheck": each mode has a pixel clock; this pixel clock is validated against the maximum pixel clock of the hardware (for a DFP, this is the maximum pixel clock of the TMDS encoder, for a CRT, this is the maximum pixel clock of the DAC). This argument disables the maximum pixel clock checking stage of the mode validation pipeline. "NoEdidMaxPClkCheck": a display device's EDID can specify the maximum pixel clock that the display device supports; a mode's pixel clock is validated against this pixel clock maximum. This argument disables this stage of the mode validation pipeline. "NoMaxSizeCheck": each NVIDIA GPU has a maximum resolution that it can drive; this argument disables this stage of the mode validation pipeline. """ And a whole bunch of others. (In reply to Elmar Stellnberger from comment #17) > By the way does anyone know whether there is a similar feature for the > radeon driver? I have an XFX Radeon R5 230 card which was advertised as 4K > ready. However I did never get any of the 4K modes to work. Wrong place to ask.
Created attachment 121499 [details] tried automatic configuration parameters: 1080p would be the maximum The automatic configuration parameters you have suggested, Ilia, and even some more of them can not replace a good manual configuration of modelines (if stated herein correctly). Nonetheless I can promise that I will give a good description on how to configure nouveau and other drivers for higher graphics modes, manually, sooner or later on my homepage which is www.elstel.org.
For now everything here has been resolved to my full satisfaction; marking as fixed. If you are ready to help me some the more pls. have a closer look at Bug 93887 (nouveau and s2ram) or Bug 93885 (radeon and TMDS overclocking). Thx!
Once more thanks a lot to everyone who has participated in extending/fixing on this issue! 2160p/UHD/4K is really a great experience especially on Core 2 / BIOS based hardware (which I personally prefer over UEFI systems).
Here are the exact steps I have taken to achieve 3840x2160 with my NVIDIA Corporation G96M [GeForce 9600M GT]: http://www.elstel.org/software/hunt-for-4K-UHD-2160p.html.en Besides this I would believe this report to be very useful for anyone else who would like to test for nouveau.hdmimhz and/or nouveau.duallink.
By now the article about 4K/UHD/2160p mentioned in Comment #22 has been released officially. This is now a fully revised version considered to be much better than the original pre-release. You may additionally view the article as translated into German. Comments as well as further usage by testers for improving the nouveau driver are highly welcome.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.