After CUDA 10.1 install on desktop (Intel i7 4790) with GTX 1070 on Ubuntu 18.04
with all updates installed, I have a black screen after log in.
Using the Driver from CUDA 10.0 (410.48) the system is working fine.
I tried the Driver 418.43 too and had the same black screen problem.
Thanks.
Please run nvidia-bug-report.sh as root and attach the resulting .gz file to your post. Hovering the mouse over an existing post of yours will reveal a paperclip icon.
https://devtalk.nvidia.com/default/topic/1043347/announcements/attaching-files-to-forum-topics-posts/
Hi,
I solved the problem with a fresh install on Ubuntu and install the latest driver and install the cuda(.run) file and not replacing with the 418.39. But if I run the .deb it replaces the driver and the black screen continues. The .gz is attached.
Thanks
nvidia-bug-report.log.gz (1.12 MB)
Hi,
I am having the same issue of Black screen after installing cuda-10.1 with a deb(network) file.
I have not tried a run file install.
Cuda 10 with 410 driver was working fine on my system till the update on February 28. Currently cuda is not installed on my system.
I have desktop running 18.04, i7 8700k and GTX 1080 Ti.
Hi,
I installed the driver 418.43 and after that run the cuda (.run) file where it
prompts me what to install, so I chose to not install the driver 418.39.
It is working fine now.
I had a similar issue about this and here is how I walked through the problem and eventually made it. I have 2080 Ti and Ubuntu 18.04. I have two hard drives and have Ubuntu and Windows 10 installed on separate drives.
I iteratively tried out version 418, 415 and 410 without reinstalling the system and none of them worked. I had the 410 worked on my machine before (I installed that through the run file from CUDA 10). Later I realized it could be the problem appeared in this post (https://askubuntu.com/questions/61396/how-do-i-install-the-nvidia-drivers/680826) so I went to change the WaylandEnable parameter and even disable secure boot and still, nothing worked. However I found out the problem could be coming from the UEFI mode.
Finally I re-installed my 18.04 through a bootable USB. The catch here is to choose boot from regular USB boot, NOT UEFI USB mode, in the BIOS. I didn’t pay attention to it when installing it first and that caused the problem. Later I just downloaded the CUDA 10.0 run file again and followed through the driver and CUDA installation and it finally worked. I didn’t dare to try the 418.xx version driver but chose 410.48.
Hello,
I have dual boot of ubuntu 18.04.2 and windows 10.
Instead of ubuntu desktop installed Ubuntu server with the alternative ubuntu server installer.
Installed nvidia driver 415.27 from the ppa repository.
Installed cuda-toolkit-10-0 meta package with .deb(network) installer.
Tested with cuda samples, tensorflow and pytorch.
Hi,
I suffered the exact same problem. Freshly installed Kubuntu 18.04.2, then CUDA 10.1 with the .deb (network) option. 4790K with GTX 970.
I see that installing NVIDIA driver 418.43 with the .run file, then CUDA 10.1 with the .run file omitting the driver part seems to work, but I’d rather stick with a solution with package manager support.
For now I’ll install the latest NVIDIA driver offered in the official distribution repository and then the meta package cuda-toolkit-10 from the CUDA 10 .deb (network) install, until the latest offered driver exceeds 418.39 (since this seems to be the culprit), when I’ll move to cuda-toolkit-10-1.
Hello
On my dual boot system Ubuntu 1804/Win10 (1080TI) I suffered from the same black screen problem.
After bringing back the driver to 410 I was able to boot normal and compile with nvcc 10.1.
I was however not able to run the cuda 10.1 samples (wrong driver of course). After replacing the driver with the new driver NVIDIA-Linux-x86_64-418.43.run everything seems to work ok under cuda 10.1.
Same issue here. Ubuntu 18.04 with duo 1080Ti. The login screen is fine, but it goes all black after login. Several facts observed:
The 418.43 driver is now in the Ubuntu graphics ppa. Install that then install just the toolkit: apt install cuda-toolkit-10-1
The 418.43 driver packaged in the repository simply does not work.
As of today, on Ubuntu 18.04.2 LTS, using a Titan X (Pascal) set up which has been functional for 2 years, after a:
sudo apt purge ‘nvidia*’ ‘cuda*’
sudo apt autoremove
sudo apt autoclean
sudo apt clean
and even a reinstall of the kernel 4.15.0-46
Installing the nvidia-driver-418 and restarting the machine creates the same black screen after the login page.
As other users here, switch to other tty is not possible. Only remote ssh connection allows operations on the machine.
After another uninstall everything works fine with opensource drivers.
Can Nvidia please provide clear steps on how to solve the issue?
Thank you
If you can login via ssh when the black screen is there, please run nvidia-bug-report.sh as root and attach the resulting .gz file to your post. Hovering the mouse over an existing post of yours will reveal a paperclip icon.
https://devtalk.nvidia.com/default/topic/1043347/announcements/attaching-files-to-forum-topics-posts/
I just tried to reinstall the driver to get the logs as requested and got a strange behavior.
During the time necessary to get the logs in the black screen, the loging page popped again and I could try to login again.
Then, after a new black screen, I waited and finally logged in.
Now I tried to do a nvidia-smi:
Sun Mar 10 17:37:40 2019
±----------------------------------------------------------------------------+
| NVIDIA-SMI 418.43 Driver Version: 418.43 CUDA Version: 10.1 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 TITAN X (Pascal) Off | 00000000:01:00.0 On | N/A |
| 23% 45C P0 59W / 250W | 172MiB / 12192MiB | 0% Default |
±------------------------------±---------------------±---------------------+
±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 3176 G /usr/lib/xorg/Xorg 40MiB |
| 0 3229 G /usr/bin/gnome-shell 53MiB |
| 0 3423 G /usr/lib/xorg/Xorg 40MiB |
| 0 3577 G /usr/bin/gnome-shell 34MiB |
±----------------------------------------------------------------------------+
And please find the bug report:
nvidia-bug-report.log.gz (1.12 MB)
Now I tried to install CUDA 10.1 from the repo on top of it and after restart, the black screen is permanent.
Also, here is the nvidia-smi:
Sun Mar 10 18:07:48 2019
±----------------------------------------------------------------------------+
| NVIDIA-SMI 418.43 Driver Version: 418.43 CUDA Version: 10.1 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 TITAN X (Pascal) Off | 00000000:01:00.0 On | N/A |
| 23% 34C P8 11W / 250W | 161MiB / 12192MiB | 0% Default |
±------------------------------±---------------------±---------------------+
±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 1837 G /usr/lib/xorg/Xorg 105MiB |
| 0 2105 G /usr/bin/gnome-shell 53MiB |
±----------------------------------------------------------------------------+
Please find the new debug log file after install of CUDA:
nvidia-bug-report.log.gz (1.13 MB)
Ok, looks like it’s still the same error as with the 418.39 driver from cuda package:
(EE) NVIDIA(GPU-0): Failed to acquire modesetting permission.
So it seems the .43 driver fixes the issue for most people but not for all. Since you were able to log in the second time, seems to be some kind of timing bug.
My nvidia-bug-report.log shows the same error:
[ 20.696] (II) NVIDIA GLX Module 418.43 Tue Feb 19 01:05:57 CST 2019
[ 20.697] (–) NVIDIA(0): Valid display device(s) on GPU-0 at PCI:66:0:0
[ 20.697] (–) NVIDIA(0): DFP-0
[ 20.698] (–) NVIDIA(0): DFP-1
[ 20.698] (–) NVIDIA(0): DFP-2
[ 20.698] (–) NVIDIA(0): DFP-3 (boot)
[ 20.698] (–) NVIDIA(0): DFP-4
[ 20.698] (–) NVIDIA(0): DFP-5
[ 20.698] (–) NVIDIA(0): DFP-6
[ 20.699] (II) NVIDIA(0): NVIDIA GPU GeForce GTX 1080 Ti (GP102-A) at PCI:66:0:0
[ 20.699] (II) NVIDIA(0): (GPU-0)
[ 20.699] (–) NVIDIA(0): Memory: 11534336 kBytes
[ 20.699] (–) NVIDIA(0): VideoBIOS: 86.02.39.00.bd
[ 20.699] (II) NVIDIA(0): Detected PCI Express Link width: 16X
[ 20.699] (EE) NVIDIA(GPU-0): Failed to acquire modesetting permission.
[ 20.699] (EE) NVIDIA(0): Failing initialization of X screen 0
[ 20.701] (II) UnloadModule: “nvidia”
[ 20.701] (II) UnloadSubModule: “glxserver_nvidia”
[ 20.701] (II) Unloading glxserver_nvidia
[ 20.701] (II) UnloadSubModule: “wfb”
[ 20.701] (II) UnloadSubModule: “fb”
[ 20.701] (EE) Screen(s) found, but none have a usable configuration.
[ 20.701] (EE)
Fatal server error:
[ 20.701] (EE) no screens found(EE)
[ 20.701] (EE)
Hi,
I went through the above discussion and conclude that installing cuda 10.1 using deb (network ) causing black screen.
But I will really appreciate if someone can confirm exact repro steps along with nvidia bug report so that I can try it internally.
Thanks in advance.
It might be interesting to know if this is only hapening on Xserver 1.19, can you upgrade to 1.20 by upgrading the HWE stack?
https://wiki.ubuntu.com/Kernel/LTSEnablementStack