331.20 WHQL long-term driver discussion



ERROR: The Unified Memory kernel module failed to build. To work around this issue, you may attempt to install this driver package again with the '--no-unified-memory' option.

This is on 3.12 64 bit, 331.20 driver. It does not install without that option on that system, but I’m not clear what the negatives are of forcing that option on the user systems in general, there is no real data about it. 319.72 installs fine on that same 3.12 kernel, not having the nvidia-uvm.ko module I guess.

Nor is there any meaningful explanation about why some users have no problem building 331. drivers without this option, and some do. Usually build failures affect all systems, or at least most.

I’ll see if I can find get some users to supply nvidia bug report files on failed installs, and ideally, on non failed installs with 3.12 kernels.

This module is currently not being used by anything, so sleep tight.

I successfully installed the 331.20 driver via repo. It fixed the compatibility issues with the kernel 3.11, but I still have problems after resuming from sleep mode: the screen appears distorted and nothing is readable, but I can still type ALT+F2 kwin --replace and everything come back. Sometimes it happens that the screen is completely black except for a gray stripe at the top of it. In this case the previous trick doesn’t work and I have to switch to a virtual terminal and reboot.
I use Kubuntu 13.10. I asked help to KDE support team and they answered that it’s likely a driver bug. Any idea?

ps: Is this the right thread for this issue or should I start a new one?

I installed the version in xorg-edgers/ppa on Ubuntu 13.10 without needing any patches on my desktop and laptop. I had to upgrade the kernel version on my laptop to 3.12 from 3.12-rc3 however, because the driver would not build the kernel module under the RC kernel… so make sure you upgrade to the 3.12 kernel release before you install the new driver.

Hi, people.

I was having similar problems like _tomas, the dkms module wouldn’t build. I applied the patch birdie provided, and now I get several warnings, but the module compiles. However, ‘modprobe nvidia’ now says “ERROR: could not insert ‘nvidia’: Exec format error”. Now what? :/

I’m using Debian Sid with freshly compiled 3.12 kernel. Hardware: Intel Core i7 + GeForce 650M on an Optimus notebook.

The warnings I mentioned… WARNING: could not find /var/lib/dkms/nvidia/331.20/build/.nv-kernel.o.cmd for / - Pastebin.com

Thank you in advance.

I just replaced euid with euid.val
Should I better use your version?

I installed the version in xorg-edgers/ppa too, so I didn’t need to compile it. Though I remember that I had problems with dkms, but they didn’t show by installing via repo. I didn’t try to install the kernel 3.12, because I’m not so confident with that kind of stuff and I don’t want to screw up the system by messing up with kernels…

most nvidia-settings queryable software attributes not view because don’t have target



nvidia-settings -q OpenGLVersion

ERROR: Error resolving target specification '' (No targets match target specification), specified in query 'OpenGLVersion'.

(tested with :0 , :0.0, sL1pKn07:0.0…)

and other duplicate with and without atribute:

Attribute 'ConnectedDisplays' (sL1pKn07:0.0): 0x00010000.
    'ConnectedDisplays' is a bitmask attribute.
    'ConnectedDisplays' is a read-only attribute.
    'ConnectedDisplays' can use the following target types: X Screen, GPU.
Attribute 'ConnectedDisplays' (sL1pKn07:0[gpu:0]): 0x00010000.
    'ConnectedDisplays' is a bitmask attribute.
    'ConnectedDisplays' is a read-only attribute.
    'ConnectedDisplays' can use the following target types: X Screen, GPU.

This driver doesn’t seem to have fixed the “gpu has fallen off the bus” problem. The only kernel that I have installed that will actually boot is 3.9.8 (Debian 3.9-1). All of the others, including 3.10, various 3.11 kernels, and 3.12 fail to bring up the gpu with this version of the driver.

[This file was removed because it was flagged as potentially malicious] (43.9 KB)


seems to the the kernel compile option that is causing the failures, at least that’s what initial tests have shown. This would explain as well why some kernels failed and some worked, it all may just depend on the person who is creating the config options for the compile.

This is just noted FYI for nvidia devs.

We’re trying this in that case, search /boot/config-$(uname -r) for the compile option, if present, use the nvidia installer option: --no-unified-memory

which was the early reported ‘fix’ for this issue.

We’ll see on this forum thread how it goes.