Driver 364.19: XServer 1.18.3 crashes - not enough displayport bandwidth

System Information:

  • Ubuntu 16.04 with latest updates
  • NVIDIA driver 364.19 downloaded from NVIDIA and installed with the official installer
  • XServer Version: 2:1.18.3-1ubuntu2
  • Monitors: 2 times DELL U2715H (2560x1440) connected via Display Port
  • Graphics Card: NVIDIA Corporation GM206 [GeForce GTX 960] (rev a1)

My Xserver crashes very often after resuming from suspend-to-ram. I see a black screen for several seconds (around 10-15sec.) and then I get back to the login screen. This happens more then 50% of all resumes.

When I look at the

Xorg.0.log.old

I find the following output just before the crash:

[ 90494.481] (II) NVIDIA(0): Setting mode "DP-4.8: nvidia-auto-select @2560x1440 +2560+0 {ViewPortIn=2560x1440, ViewPortOut=2560x1440+0+0}, DP-2.8: nvidia-auto-select @2560x1440 +0+0 {ViewPortIn=2560x1440, ViewPortOut=2560x1440+0+0}"
[b][ 90494.481] (WW) NVIDIA(0): Failed to set the display configuration
[ 90494.481] (WW) NVIDIA(0):  - Setting a mode on display 0 failed: There was insufficient
[ 90494.481] (WW) NVIDIA(0):     available DisplayPort bandwidth
[ 90494.481] (EE) NVIDIA(0): Failed to enter VT (mode initialization failed)[/b]
[ 90494.481] (EE) 
Fatal server error:
[ 90494.481] (EE) EnterVT failed for screen 0
[ 90494.481] (EE) 
[ 90494.481] (EE) 
Please consult the The X.Org Foundation support 
         at http://wiki.x.org
 for help. 
[ 90494.481] (EE) Please also check the log file at "/var/log/Xorg.0.log" for additional information.
[ 90494.481] (EE) 
[ 90497.178] (EE) Server terminated with error (1). Closing log file.

nvidia-bug-report.log.gz (218 KB)
Xorg.0.log.old.tar.gz (5.92 KB)

Dear NVIDIA devs,

it has been almost one month since I reported this and there has been zero response. Do you need more information? Is there anything I can do to avoid these crashes? Is there a known workaround?

I am currently getting these crashes everyday multiple times. As I wrote in the original post, this happens more then 50% of the times when I suspend, and I usually suspend my computer several times a day. This means that I loose my session and all opened programs at least once per day! I find this unacceptable!

The graphics card I am using cost me 200 Euro! I expect to get some value for this price and not a system which keeps crashing all the time! Please fix this!

Nvidia doesn’t have Linux dev’s. Their driver doesn’t even run on the stable 4.6 kernel. And everyone here wipes with their left hand, and don’t know a kernel from a corn cob. So I suggest getting used to disappointment and work around their stupidity.

Just save your money for AMD’s RX4xx. They are developing kernel native drivers.

Its still crashing with the latest update 367.27. And still no reply from NVIDIA.

When you resume your computer from suspend, are your monitors already powered on and in standby mode, or are you powering them on and resuming the computer at roughly the same time?

This problem can occur if link training fails, which can happen if the monitor is still booting when you try to train the link. If there’s not enough bandwidth, the alternatives aren’t great – really all we can do is just not enable the displays, which is also a pretty terrible experience.

Dear Aaron,

thank you for the response. I do not power off the monitors when I enter suspend, they are always switched on. When the PC enters the suspend state I see a brief message that shows “Entering power save mode” and the screens turn dark.

I do not really understand how there cannot be enough bandwidth since during normal operation I do not have any problem with my NVIDIA card nor the monitors. Would be possible to delay the link training for a brief moment or make that delay configurable somewhere? Or is there a way to detect when the link has become ready?

There is one more information that I can provide which may be related. When I use the shortcut CTRL+ALT+F1 to switch to TTY1 then I can observe a similar behaviour. The screen becomes dark and I can see the “Entering power save mode” for a short moment then it stays dark for about 4 seconds until the image comes back and I see the login prompt. When I switch back to the X server on TTY7 I see the same phenomenon, but it only takes 2 seconds.

Occasionally I also see that one of my screens turns back for no apparent reason, and the other continues to show an image. I can still interact with the second monitor, so that means its not a problem with the X server. Usually the image the comes back after a 2-3 seconds, but sometimes I need to switch to TTY1 and back to restore the image on that monitor.

On my notebook with Intel Graphics switching to TTY1 is almost instantaneous.

One more idea. If link training fails, would it be possible to enter a retry loop that tries link training a few more times until it fails? If your assumption that the monitor is still booting (even if its just resuming from power save mode, then it should usually work on the second or third retry, right?