1

I have a Lenovo Thunderbolt 3 Graphics Dock. This is based on a NVIDIA GeForce GTX 1050 graphics card. lspci says:

07:00.0 VGA compatible controller: NVIDIA Corporation GP107M [GeForce GTX 1050 Mobile] (rev a1)

It works with nvidia-driver-390 using the proprietary ppa:

enter image description here

However, when I upgrade to a higher NVIDIA-Version then the system does not even boot up. I tried nvidia-driver-410, nvidia-driver-430 nvidia-driver-435, nvidia-driver-440 (I did a sudo .pt purge 'nvidia.*' before that).

How come? According to the Supported Products tab the GeForce GTX 1050 should work with e.g. nvidia-driver-430. In fact, it works without problems in Windows 10 (I have a dual boot setup).

Since it does not even boot, I have no idea to debug this (any logs?) or how to track this down. Also where would I report such bugs?

Ubuntu 19.10 with Gnome and Xorg.

  • Following, I have a 1050 Ti and I've been afraid to go to a newer driver for just this reason. – Organic Marble Jan 8 at 16:50
0

When installing graphics drivers, I too find that I do not know which driver to install. I find that the safest way to install NVidia drivers is by, in fact, installing CUDA instead. When you install CUDA, it brings with it a compatible graphics driver.

If you select CUDA 10.1, for example, it will bring proprietary graphics driver version 4.18 with it.

I also surmise, given my experience, that components like cooling fans and temperature sensors do NOT load properly without CUDA installed. (you may have experienced differently, however) Basically, I do not trust any other graphics driver that does NOT come with CUDA.

My suggestion, if you are trying to install the most recent driver safely, would be to install the most recent version of CUDA, which, at the time of writing this, is version 10.2.

There are several ways to install CUDA, but I will narrow down the choices so that the decision making is easier.

  1. Local Deb file installation
  2. Local Run file installation

The first option is the best option for people that decide they don't want to configure anything and just want their machine to work. You will not know where anything is installed since deb installations follow the OSs decisions.

The second option is best when you want to configure where libraries are installed and other extra options, which, in this context, are irrelevant. This installation allows you to choose what features to keep. In fact, you could install CUDA without a driver, but that assumes you have your environment setup exactly how it is required by NVidia.

I suggest you take the first option so that you don't brick your computer.

I am assuming that you are running Ubuntu 1804 in this case.

What you will want to do is go to this link and then select Linux, x86_64, and then Ubuntu and then 1804. You will then select deb (local). A list of commands will be listed for you. I have copied and pasted them here:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb
sudo apt-key add /var/cuda-repo-10-2-local-10.2.89-440.33.01/7fa2af80.pub
sudo apt-get update
sudo apt-get -y install cuda

Here is how to start your BROKEN Ubuntu machine:

  1. Press the power button.
  2. Find the correct OS option in Grub2
  3. Go down 1 option for Advanced Options
  4. Select the first option, from the top, that says Recovery Mode. You will be taken to a weird pink background window.
  5. Select fsck to mount the OS drive.
  6. Select Enable Networking (this may not work, it did not for me, but usually does work)
  7. Then select Drop to root recovery prompt

If you cannot use a local Ubuntu machine WITH INTERNET, simply download the required files by pasting the links into a Windows machine's browser and download them.

You can then put those 2 files onto a flash drive and connect it to your broken Ubuntu machine.

Now you have a root terminal you can do stuff with. I then mounted the thumb drive through CLI following this link.

https://vitux.com/how-to-manually-mount-unmount-a-usb-device-on-ubuntu/

You can then move the 2 files over to the local computer and run them.

BTW, I know there is a sudo apt update command there, but apt should still detect new sources added.

After following those instructions, you should reboot your machine and then try to log in normally.

If you do have internet, you can simply run the commands after enabling networking.

If the deb-file install does not work, you can simply download the run file version and install in root recovery prompt following the same methods.

You should have a working graphics driver AS WELL AS CUDA after going through all of these steps. Slight modifications were added in case you don't have internet, but hopefully that is not the case. I happened to not have internet last I installed CUDA, so yeah, try that.

ONE MORE THING:

Sometimes, if you are in the broken login screen (after you typed the password and hit ENTER), there is actually a sneak you can do.

Reboot your machine but DO NOT log in. You may also want to check if you have internet. Now, instead of normally logging in, press Ctrl+Alt+F3 to go to a virtual serial port. You should then be able to log in through the terminal. You should be able to have access to the internet. At this point, you can simply run the original commands as before without changes.

Does this help you?

|improve this answer|||||
  • I should also mention: since you are using a dock to get the GPU, you need to install these drivers WHILE CONNECTED TO THE DOCK. Then, since you will loose the GPU when you undock, you need to be able to go back to non-NVidia drivers (assuming you don't have an onboard GPU). This means that you will need to have a different graphics driver that is NOT blacklisted. The CUDA installation manual talks about blacklisting Nouveau drivers somewhere in the manual. You may need to check to make sure your alternatives are not blacklisted. – thatrobotguy Jan 8 at 22:08
  • Thanks. Actually I have an on-board intel on the Thinkpad. So I do not need the "unbrick" help. But I'll try to use CUDA and see if this helps. I doubt a bit because I've already tried all nvidia-driver-* and all fail except nvidia-driver-390 thought they should support my card. Also, I'm on 19.10 - do you think CUDA for 18.04 also works for 19.10? – nachtigall Jan 9 at 12:19
  • I know what you mean: nvidia-driver-* should just work, right? Except no, some of those are open-source drivers. Those don't always work. I find that the ppas are often too out of date to work, even if they are the proprietary ones. The thought process for me was "If CUDA knows of a good driver, let it choose which one to use, not the user". That means you don't have to figure out which one works, it just gives you the right one. – thatrobotguy Jan 9 at 16:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.