NVidia driver often crash on my Ubuntu 16.04

Question

I have two NVidia GPU on two PCs: a GTX1080 and a GTX1050. Both OS are Ubuntu 16.04LTS and have been updated in the last month with the latest drivers from the official website installed. I often observe what I think is driver crash: I get stuck in the log-in loop in the GUI (login screen-> type password and Enter -> black out->login screen again) and the python-lasagne GPU backend cannot start. By reinstalling the driver, everything comes back alive right away.

The GTX1080 machine is my daily work machine and its driver crash quite often: whenever I unplug/plug-in on of my dual-monitors when the PC is down, I will surely get the crash. Today I replaced one of my monitors and it happened again.

The GTX1050 machine is used as a server and when we had an unexpected power-off, the driver issue came out (and got fixed by the same way).

My question is: is this normal? Can I avoid it some way or at least get it fixed faster than reinstalling the whole standard+cuda driver (not that slow but annoying)? Both machines have Win10 installed and never had no problem on the video card, so I suppose this is a Ubuntu bug.

Answer 1

If you solved the login loop without reinstalling the OS please post your solution. We also had it when preparing Ubuntu 16.04.2 for theano - lasagne. The cocktail that worked for us is to: 1. switch to Nvidia driver in "additional drivers" (obvious but has to be stated) 2. follow Theano's installation instructions for Ubuntu. We installed Anaconda rather than the suggested miniconda. Install the dependencied there. Note that Theano itself should be installed following Lasagne guidlines 3. Follow lasagne installation guidlines, including installation of recent theano pack and not the default for Ubuntu.

Answer 2

I have two pc's one with nvidia old 9600 GT and one more modern with nvidia GT 440 card. Drivers used for the older Nvidia drivers 340 non free (can't go higher since then it crashes) Drivers on the newer pc Nvidia 384 non free. But sometimes the gnome3 desktop crashes. 3 out of 5 times the desktop recovers and a error report is sent to ubuntu. 2 times a reboot is needed. And from that reboot (using pc reset) 1 times out of two restart of pc impossible. (with gdm3) no login screen appears with lightgdm login appears but in low resolution. If you try to login new crash occurs. Only solution then is using ubuntu upstart in boot menu which just start ubuntu without graphical support. then using apt-get remove purge nvidia* this followed by apt-get install nvidia-340 (for the old since that's the driver version used) nvidia-384 for the modern pc. After that all runs fine again. This problem came up the day gnome3 started to be used (with unity however it's the same issue). Somehow under certain situations I suspect that a segfault occurs, unfortunately I never could find how and why. It is completely at wild which is well typicall for a segfault. I do not think it is related to the video card self but looks to be a problem from nvidia driver together with gnome3 and under some circumstances. The worsted of it is the requirement to uninstall (and purge !) the nvidia driver, looks like up on crash one of the settings is screwed sometimes. It is very weird, but it is with two different pc's. One older: Intel® Core™2 Duo CPU E8600 @ 3.33GHz × 2 , GeForce 9600 GT/PCIe/SSE2( Gigabyte P-45 extreme MB), 4 GB ram. One newer Intel core i7 cpu x 990 3.47Ghz x 12 , Geforce GT 440/PCIe/SSE2 (asus P6 X58 Ws MB), 12 GB RAM .

Both with ubuntu 16.04 Lts Gnome3 desktop.