Titan V, Ubuntu 16.04LTS and 387.34 driver crashes badly

Just built new computer with Titan V. Have set up for Windows and set up for Ubuntu. Under Windows, the driver is mildly buggy - causing Matlab to hang for three minutes when trying to access the GPU, but ultimately working well. Under Ubuntu, it’s a complete failure for me (and this is the configuration I really want to run). Matlab completely hangs when attempting to access the GPU. Doesn’t leave any diagnostics that I can find. VMWare completely hangs the X-windows server when it starts up after installing the driver. Again, no error messages or logs that I can find. What can I do to get this running, please? I feel like I’ve just wasted $3k on a card I really can’t access.

What kind of system setup are you running?
Please run nvidia-bug-report.sh and attach output tar.gz file to your post.

Actually, it is worse than I thought. Just to make it simple, I did an absolute clean install of Ubuntu 16.04 LTS. Then did sudo apt update, sudo apt upgrade, sudo shutdown -r now

So that is an absolutely clean install, with all latest patches. Downloaded driver, followed instructions for installation, rebooted. And lots and lots of different things caused hang/crash. Firefox. Term. Others did not. Finally, ended up calling up virtual terminal via ctrl-alt-F1, logging in, running command from that terminal, mounting one of the Windows drives and moving the file over.

Yeah, the driver for Ubuntu 16.04 seems broken pretty hard. Don’t see a mechanism for adding a file to the post here. Please send me a message as to how to get the file to you.

Actually, it is worse than I thought. Just to make it simple, I did an absolute clean install of Ubuntu 16.04 LTS. Then did sudo apt update, sudo apt upgrade, sudo shutdown -r now

So that is an absolutely clean install, with all latest patches. Downloaded driver, followed instructions for installation, rebooted. And lots and lots of different things caused hang/crash. Firefox. Term. Others did not. Finally, ended up calling up virtual terminal via ctrl-alt-F1, logging in, running command from that terminal, mounting one of the Windows drives and moving the file over.

Yeah, the driver for Ubuntu 16.04 seems broken pretty hard. Don’t see a mechanism for adding a file to the post here. Please send me a message as to how to get the file to you.
nvidia-bug-report.log.gz (65.8 KB)

Hover the mouse above an existing post, in the right upper corner of it icons appear for editing and attaching files.

Yeah. I figured that out. Please see previous post. I’d love to get a solution to this one…

FYI: ASROCK Taichi X299 motherboard. I9 7900X. 64 GB ram. 2 M2 1TB SSDs, a few hard drives. Titan V GPU. And that’s it. For testing, all UEFI defaults, no overclocking of any kind. Trying to keep it as simple as possible. (Ubuntu installed on one of the hard drives, fully utilized. All other drives set up for Windows - I still have to do real work with this thing).

Unfortunately, no errors are logged. Did you run nvidia-bug-report.sh right after crashes occured without reboot?
Did you use the graphics ppa to install the driver?

Okay. Tried this. Installed openssh-server. Logged in remotely from another Linux box using ssh. Logged in (graphically) to the box with the Titan V. Opened Firefox. Hangs with spinning wheel of wait. Watched processes from remote Linux box - Xorg continuously chewing up 100% CPU. Wait, wait, wait, nothing happens. Okay, ran the bug report. I doubt it will show a crash, since Xorg is hanging. A hint may be that the local error file shows that the system cannot connect to br1tty

Also, in going through the logs, it seems that the driver doesn’t really recognize the Titan V as an NVidia card?
nvidia-bug-report.log.gz (68.8 KB)

And sorry, I used this to install the drivers (as indicated at the driver download page):

i) dpkg -i nvidia-driver-local-repo-ubuntu1604-387.34_1.0-1_amd64.deb’ for Ubuntu ii) apt-get updateiii)apt-get install cuda-drivers iv)reboot`

I’ve now tried clean installs of Unbuntu, RHEL, Fedora, CentOS. Not a single one of them works with the NVidia driver. Every last one of them crashes/freezes (except for RHEL - where I couldn’t install dkms, so I couldn’t install the driver).

Maybe run some Unigine benchmarks under Windows as a general system test and ask these people
[url]Nvidia-smi not recognizing Titan V - Linux - NVIDIA Developer Forums
how they’re getting along with their Titan V.

Thanks for the additional link. I’ve been benching with Sandra and also Matlab (since I’m using the card with Matlab, that last one makes a lot of sense). Initial startup of the card is horribly slow under Windows - 3 full minutes. But once that’s done, it routinely hits >4 TFlops double precision, which is what I bought it for. It’s just that Windows is an incredibly painful environment for me to work in (data handling stuff), and I REALLY want to run this thing under Ubuntu…

There’s something very wrong either with your hardware setup or the drivers. 3 minutes init time doesn’t sound right. Did you check if it works in another slot? If all fails, try to use the iGPU for graphics and the Titan as mere accelerator for a workaround.

  • turn on iGPU in bios, connect your monitor to that
  • clean install ubuntu
  • use the .run installer with option --no-opengl-files

It’s the driver. Plugged in a Titan Black. With earlier drivers, it all works fine. With the latest driver, it croaks. Sorry, guys.

I would try updating motherboard BIOS and using another PSU. Also, try disabling any OC’ing if you have one (CPU/RAM/PCI-E). I would even recommend resetting BIOS settings to safe defaults.

Titan V is quite a beast which warrants top quality components.

Already updated BIOS, disabled all OC, cleared CMOS to basic settings. Tried swapping PSU. Symptoms remain the same. Windows is rock solid. (Mathworks has confirmed that the initial delay is to be expected - it takes them that long to compile and load a binary to the GPU when it’s a model they don’t have in their database). Linux is flaky. New datum - Linux driver frequently returns NaN as available GPU memory on first attach. Windows driver does not exhibit this behavior. I’m going to try swapping slots for the GPU. I can’t use built-in graphics for display - the motherboard doesn’t have it. Seems silly to do that when the Windows system works just fine. Switching slots may be a good thing to try. I may even swap it into a different computer just as a driver check.

390 beta driver is out renewing Titan V support.

I’m not finding that beta driver. Can you provide a link? It’s not showing up on the Advanced Driver Search when looking for English language, Ubuntu 16.04 64-bit, Titan V, beta

[url]Current graphics driver releases - Linux - NVIDIA Developer Forums