"sudo apt update" not working anymore after installation of CUDA Toolkit 13.0 Update 2 Downloads

stefan_bohl · November 21, 2025, 7:13pm

Hello, i have done a mistake with a installation on my new DGX sparc

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/cuda-ubuntu2404.pinsudo mv cuda-ubuntu2404.pin /etc/apt/preferences.d/cuda-repository-pin-600wget https://developer.download.nvidia.com/compute/cuda/13.0.2/local_installers/cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.debsudo dpkg -i cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.debsudo cp /var/cuda-repo-ubuntu2404-13-0-local/cuda-*-keyring.gpg /usr/share/keyrings/sudo apt-get updatesudo apt-get -y install cuda-toolkit-13-0

now i get this error

E: Conflicting values set for option Signed-By regarding source Index of /compute/cuda/repos/ubuntu2404/sbsa /: /usr/share/keyrings/cuda_debian_prod.gpg != /usr/share/keyrings/cuda-archive-keyring.gpg
E: The list of sources could not be read.
ma10a07@spark-9452:~$

also the app software updater is not working anymore

What can i do, it the only way the recover procedure?

_cjg · November 21, 2025, 11:28pm

Is this your keyring?

$ wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/cuda-keyring_1.1-1_all.deb
$ sudo dpkg -i cuda-keyring_1.1-1_all.deb

from this page?

stefan_bohl · November 22, 2025, 9:05am

Yes, it is from this page

user@spark-9452:~$ sudo dpkg -i cuda-keyring_1.1-1_all.deb

[sudo] password for user:

(Reading database … 322074 files and directories currently installed.)

Preparing to unpack cuda-keyring_1.1-1_all.deb …

Unpacking cuda-keyring (1.1-1) over (1.1-1) …

Setting up cuda-keyring (1.1-1) …

user@spark-9452:~$ wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/cuda-ubuntu2404.pin

--2025-11-22 18:21:37-- https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/cuda-ubuntu2404.pin

Resolving developer.download.nvidia.com (developer.download.nvidia.com)… 2.16.241.198, 2.16.241.216

Connecting to developer.download.nvidia.com (developer.download.nvidia.com)|2.16.241.198|:443… connected.

HTTP request sent, awaiting response… 200 OK

Length: 190 [application/octet-stream]

Saving to: ‘cuda-ubuntu2404.pin’

cuda-ubuntu2404.pin 100%[===================>] 190 --.-KB/s in 0s

2025-11-22 18:21:37 (65.8 MB/s) - ‘cuda-ubuntu2404.pin’ saved [190/190]

user@spark-9452:~$ sudo mv cuda-ubuntu2404.pin /etc/apt/preferences.d/cuda-repository-pin-600

user@spark-9452:~$ wget https://developer.download.nvidia.com/compute/cuda/13.0.2/local_installers/cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.deb

--2025-11-22 18:21:59-- https://developer.download.nvidia.com/compute/cuda/13.0.2/local_installers/cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.deb

Resolving developer.download.nvidia.com (developer.download.nvidia.com)… 2.16.241.216, 2.16.241.198

Connecting to developer.download.nvidia.com (developer.download.nvidia.com)|2.16.241.216|:443… connected.

HTTP request sent, awaiting response… 200 OK

Length: 3838491326 (3.6G) [application/x-deb]

Saving to: ‘cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.deb.2’

cuda-repo-ubuntu240 100%[===================>] 3.57G 16.1MB/s in 4m 6s

2025-11-22 18:26:05 (14.9 MB/s) - ‘cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.deb.2’ saved [3838491326/3838491326]

user@spark-9452:~$ sudo dpkg -i cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.deb

(Reading database … 322074 files and directories currently installed.)

Preparing to unpack cuda-repo-ubuntu2404-13-0-local_13.0.2-580.95.05-1_arm64.deb …

Unpacking cuda-repo-ubuntu2404-13-0-local (13.0.2-580.95.05-1) over (13.0.2-580.95.05-1) …

Setting up cuda-repo-ubuntu2404-13-0-local (13.0.2-580.95.05-1) …

user@spark-9452:~$ sudo cp /var/cuda-repo-ubuntu2404-13-0-local/cuda-*-keyring.gpg /usr/share/keyrings/

user@spark-9452:~$ sudo apt-get update

E: Conflicting values set for option Signed-By regarding source Index of /compute/cuda/repos/ubuntu2404/sbsa /: /usr/share/keyrings/cuda_debian_prod.gpg != /usr/share/keyrings/cuda-archive-keyring.gpg

E: The list of sources could not be read.

user@spark-9452:~$

_cjg · November 22, 2025, 6:59pm

This is really stupid, because it also kills $ apt-get update. This doesn’t help you directly, but maybe it narrows down the problem a bit.

stefan_bohl · November 23, 2025, 10:57am

Hi,

so what i could do now, any suggestions, to get out?

_cjg · November 23, 2025, 11:22am

Sorry, my expertise isn’t extensive enough to resolve this issue. I have a theory, but NVIDIA Support should address it. It can’t be that their own instructions for NVIDIAs glorified CUDA broke apt on the Spark… well, it’s the weekend, and they’re probably just sitting in Verve/PA, which they deserve after this week ;). Seriously, though, another problem is that the thread is in the project forum, where the experts, who are usually very helpful, don’t check in that often ..

cosinus · November 23, 2025, 3:19pm

You could move the conflicting .sources file from /etc/apt/sources.list.d which contains the definition of the cuda-repo.

For example: cuda-compute-repo.sources → cuda-compute-repo.sources.broken

That would allow you to use the other repos for installing software before fixing the cuda part.

The mods will probably recommend to use the official recovery procedure - reimaging the system disk with the rescue stick.

But it should be possible to recreate the needed files otherwise IMHO.

Hmm.

E: Conflicting values set for option Signed-By regarding source Index of /compute/cuda/repos/ubuntu2404/sbsa /: /usr/share/keyrings/cuda_debian_prod.gpg != /usr/share/keyrings/cuda-archive-keyring.gpg

Somewhere the original key seems be configured → /usr/share/keyrings/cuda_debian_prod.gpg is in my cuda-compute-repo.sources file.

Can you check if there are more than one .sources file in your /etc/apt/sources.list.d that have the cuda repo …/compute/cuda/repos/ubuntu2404/sbsa/?

May be you could post the content of all files (if more than one) that contain cuda in its repo path to compare those.

root@deepthought:/etc/apt/sources.list.d# grep -l cuda *
cuda-compute-repo.sources

root@deepthought:/etc/apt/sources.list.d# cat cuda-compute-repo.sources
Types: deb
URIs: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/
Suites: /
Signed-By: /usr/share/keyrings/cuda_debian_prod.gpg

And I would go for the deb (network) version as all other repos are also updated via network. If your site allows network connections to the outside world. If not - ignore this comment.

stefan_bohl · November 23, 2025, 3:56pm

Hi,

see below:

user@spark-9452:/etc/apt/sources.list.d$ grep -l cuda *
cuda-compute-repo.sources
cuda-ubuntu2404-13-0-local.list
cuda-ubuntu2404-sbsa.list
user@spark-9452:/etc/apt/sources.list.d$

user@spark-9452:/etc/apt/sources.list.d$ cat cuda-compute-repo.sources
Types: deb
URIs: Index of /compute/cuda/repos/ubuntu2404/sbsa
Suites: /
Signed-By: /usr/share/keyrings/cuda_debian_prod.gpg
user@spark-9452:/etc/apt/sources.list.d$

cosinus · November 23, 2025, 4:08pm

Those seem to be the two newly installed. If you move or rename them, you should be able to use the CUDA regular repo again.

After that I would wait for feedback of the mods on how to use the 13.0 U2 version. I would expect that the 13.0 U2 will become also available in the regular repo.

stefan_bohl · November 23, 2025, 4:15pm

So just to:

mv cuda-ubuntu2404-13-0-local.list save_cuda-ubuntu2404-13-0-local.list
mv cuda-ubuntu2404-sbsa.list save_cuda-ubuntu2404-sbsa.list

that is all ?

but it seems not so new
user@spark-9452:/etc/apt/sources.list.d$ ls -l
total 68
-rw-r–r-- 1 root root 175 Jun 26 07:57 ai-workbench-desktop.sources
-rw-r–r-- 1 root root 1866 Oct 14 21:05 canonical-nvidia-ubuntu-linux-firmware-mbssid-patches-noble.sources
-rw-r–r-- 1 root root 1857 Oct 14 21:05 canonical-nvidia-ubuntu-nvidia-desktop-edge-noble.sources
-rw-r–r-- 1 root root 153 Sep 10 19:14 cuda-compute-repo.sources
-rw-r–r-- 1 root root 108 Oct 3 09:41 cuda-ubuntu2404-13-0-local.list
-rw-r–r-- 1 root root 137 Mar 27 2024 cuda-ubuntu2404-sbsa.list
-rw-r–r-- 1 root root 179 Nov 19 20:53 dgx.sources
-rw-r–r-- 1 root root 367 Nov 15 18:24 github_git-lfs.list
-rw-r–r-- 1 root root 131 Nov 17 15:24 nvhpc.list
-rw-r–r-- 1 root root 152 Jun 26 07:57 nvhpc.sources
-rw-r–r-- 1 root root 285 Nov 15 18:24 nvidia-container-toolkit.list
-rw-r–r-- 1 root root 1863 Oct 14 21:39 nv-vulkan-desktop-ppa.sources
-rw-r–r-- 1 root root 202 Nov 15 16:40 ubuntu-esm-apps.sources
-rw-r–r-- 1 root root 206 Nov 15 16:40 ubuntu-esm-infra.sources
-rw-r–r-- 1 root root 390 Sep 13 00:15 ubuntu.sources
-rw-r–r-- 1 root root 2559 Aug 6 17:28 ubuntu.sources.curtin.orig
-rw-r–r-- 1 root root 278 Nov 21 19:24 vscode.sources
user@spark-9452:/etc/apt/sources.list.d$

_cjg · November 23, 2025, 6:18pm

After falling into the same trap, I can’t say anything about that. But deleting both files solved the problem. However, I still consider it relatively critical

a) nvcc --version does not work, which leads to the installation process being started even though the toolkit is actually already installed
b) /etc/alternatives/cuda-13/bin/nvcc --version

(the missing nvcc suggest as $ sudo apt install nvidia-cuda-toolkit, which installs very old toolkit 12.x)

stefan_bohl · November 23, 2025, 10:11pm

Hi,
i also believe that just moving two files away will not solved everything, and has left some side effects,
so still i am puzzled what is the right way to go, hope someone from Nvidia knows
because at least there should be some warning on this software page when installing not inline sw within this dgx spark machine

cosinus · November 24, 2025, 10:01am

The file extension must not be .list or .sources as apt is reading those files. So when renaming change the extension or move them to another directory like your home dir.

-rw-r–r-- 1 root root 108 Oct 3 09:41 cuda-ubuntu2404-13-0-local.list
-rw-r–r-- 1 root root 137 Mar 27 2024 cuda-ubuntu2404-sbsa.list

I assume these repos are also used by the bigger ARM based boxes. That’s why they are “not so new”.

cosinus · November 24, 2025, 10:22am

BTW: After (re)moving those to files from the CUDA 13.0 U2 installation attempt, you will also have to run an:

apt update

again to make sure the package catalog is updated.

Could you check what result you get when running:

apt policy cuda-toolkit-13-0

This is the metapackage that should take care of installing everything incl. nvcc again.

It get the following output.

root@deepthought:/etc/apt/sources.list.d# apt policy cuda-toolkit-13-0
cuda-toolkit-13-0:
  Installed: 13.0.2-1
  Candidate: 13.0.2-1
  Version table:
 *** 13.0.2-1 500
        500 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa  Packages
        100 /var/lib/dpkg/status
     13.0.1-1 500
        500 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa  Packages
     13.0.0-1 500
        500 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa  Packages

Wondering if package version 13.0.2-1 is 13.0 U2… never bothered with sub versions yet.

So if the rest of the CUDA Toolkit is affected by the previous installation attempt, I would go for a “apt install cuda-toolkit-13-0” and see what happens.

NOTE: Always have a backup readily available.

stefan_bohl · November 24, 2025, 11:06am

See below without any file movement i get this right now

user@spark-9452:/etc/apt/sources.list.d$ apt policy cuda-toolkit-13-0
E: Conflicting values set for option Signed-By regarding source Index of /compute/cuda/repos/ubuntu2404/sbsa /: /usr/share/keyrings/cuda_debian_prod.gpg != /usr/share/keyrings/cuda-archive-keyring.gpg
E: The list of sources could not be read.
user@spark-9452:/etc/apt/sources.list.d$

_cjg · November 24, 2025, 12:03pm

With file movement i get this:

g@spark-db18:~$ apt policy cuda-toolkit-13-0
cuda-toolkit-13-0:
Installiert: 13.0.2-1
Installationskandidat: 13.0.2-1
Versionstabelle:
*** 13.0.2-1 600
600 Index of /compute/cuda/repos/ubuntu2404/sbsa Packages
100 /var/lib/dpkg/status
13.0.1-1 600
600 Index of /compute/cuda/repos/ubuntu2404/sbsa Packages
13.0.0-1 600
600 Index of /compute/cuda/repos/ubuntu2404/sbsa Packages

But it is 13.0.2-1 600 vs 13.0.2-1 500 …

cosinus · November 24, 2025, 1:28pm

That’s the priority.

As you followed the instructions, you also place a pin file in /etc/apt/preferences.d

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/cuda-ubuntu2404.pin
sudo mv cuda-ubuntu2404.pin /etc/apt/preferences.d/cuda-repository-pin-600

and therefore changed the default. Which seem to be 500. You could (re)move that pin file.

Package: nsight-compute
Pin: origin *ubuntu.com*
Pin-Priority: -1

Package: nsight-systems
Pin: origin *ubuntu.com*
Pin-Priority: -1

Package: *
Pin: release l=NVIDIA CUDA
Pin-Priority: 600

For more explaination on “apt pinning” see Pinning: Apt’s Best Kept Secret - allows you to control which packages get installed (prioritized) when dealing with different repos that might contain the same packages or package names. Or which package won’t be installed at all.

And you could try a

apt-get install --reinstall cuda-toolkit-13-0

not sure if this works for metapackages (which only includes a list of packages to install and/or dependencies).

Or you could try just to reinstall the nvcc.

apt-get install --reinstall  cuda-nvcc-13-0

If that fails, I would remove all cuda* packages and reinstall cuda-toolkit-13-0. That worked at least on my other boxes without re-installing them.

But may be one of the mods have better solutions beside wiping all with the rescue image.

cosinus · November 24, 2025, 1:31pm

Without any movement the *.list files still kick in.

This is what my sources.list.d looks like for reference. No extra repos installed.

root@deepthought:/etc/apt/sources.list.d# ls -la
total 52
drwxr-xr-x 2 root root 4096 Nov 19 19:08 .
drwxr-xr-x 9 root root 4096 Sep 29 14:36 ..
-rw-r--r-- 1 root root  175 Jun 26 07:57 ai-workbench-desktop.sources
-rw-r--r-- 1 root root 1866 Oct 14 21:05 canonical-nvidia-ubuntu-linux-firmware-mbssid-patches-noble.sources
-rw-r--r-- 1 root root 1857 Oct 14 21:05 canonical-nvidia-ubuntu-nvidia-desktop-edge-noble.sources
-rw-r--r-- 1 root root  153 Sep 10 19:14 cuda-compute-repo.sources
-rw-r--r-- 1 root root  179 Nov 19 18:58 dgx.sources
-rw-r--r-- 1 root root  152 Jun 26 07:57 nvhpc.sources
-rw-r--r-- 1 root root 1863 Oct 14 21:39 nv-vulkan-desktop-ppa.sources
-rw-r--r-- 1 root root  202 Nov 19 19:08 ubuntu-esm-apps.sources
-rw-r--r-- 1 root root  206 Nov 19 19:08 ubuntu-esm-infra.sources
-rw-r--r-- 1 root root  390 Sep 29 14:49 ubuntu.sources
-rw-r--r-- 1 root root 2559 Aug  6 17:28 ubuntu.sources.curtin.orig

_cjg · November 24, 2025, 5:27pm

So, thank’s alot. At first, i will have to backup (or mv) ”2025 cuda-ubuntu2404-arm64.list as well”. Then i will, try to solve the pin topic (you have been correct, i am back on 500) and wait for the conformation (NVIDIA/mods), whether that’s it …

drwxr-xr-x 2 root root 4096 Nov 23 18:17 .
drwxr-xr-x 9 root root 4096 Sep 13 00:06 ..
-rw-r–r-- 1 root root 175 Jun 26 07:57 ai-workbench-desktop.sources
-rw-r–r-- 1 root root 1866 Okt 14 21:05 canonical-nvidia-ubuntu-linux-firmware-mbssid-patches-noble.sources
-rw-r–r-- 1 root root 1857 Okt 14 21:05 canonical-nvidia-ubuntu-nvidia-desktop-edge-noble.sources
-rw-r–r-- 1 root root 153 Sep 10 19:14 cuda-compute-repo.sources
-rw-r–r-- 1 root root 119 Okt 3 09:51 cuda-cross-sbsa-ubuntu2404-13-0-local.list_backup
-rw-r–r-- 1 root root 138 Feb 14 2025 cuda-ubuntu2404-arm64.list
-rw-r–r-- 1 root root 137 Mär 27 2024 cuda-ubuntu2404-sbsa.list_backup
-rw-r–r-- 1 root root 179 Nov 20 12:35 dgx.sources
-rw-r–r-- 1 root root 367 Nov 20 18:38 github_git-lfs.list
-rw-r–r-- 1 root root 152 Jun 26 07:57 nvhpc.sources
-rw-r–r-- 1 root root 285 Nov 20 18:38 nvidia-container-toolkit.list
-rw-r–r-- 1 root root 1863 Okt 14 21:39 nv-vulkan-desktop-ppa.sources
-rw-r–r-- 1 root root 202 Nov 20 12:49 ubuntu-esm-apps.sources
-rw-r–r-- 1 root root 206 Nov 20 12:49 ubuntu-esm-infra.sources
-rw-r–r-- 1 root root 390 Sep 13 00:15 ubuntu.sources
-rw-r–r-- 1 root root 2559 Aug 6 17:28 ubuntu.sources.curtin.orig

stefan_bohl · November 24, 2025, 9:07pm

Hi any news from NVIDIA?

Topic		Replies	Views
The repository 'https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 Release' is not signed CUDA Setup and Installation	14	10883	May 2, 2022
Updating the CUDA Linux GPG Repository Key Technical Blog	70	33910	April 4, 2024
Invalid public key for CUDA apt repository Docker and NVIDIA Docker	34	65296	July 18, 2025
Notice: CUDA Linux Repository Key Rotation CUDA Setup and Installation	30	22071	June 16, 2023
18.04 CUDA docker image is broken CUDA Setup and Installation	17	25144	June 28, 2022
How to downgrade cuda? CUDA Setup and Installation	3	8598	April 17, 2023
Apt-get update is failing CUDA Setup and Installation	12	666	July 29, 2025
Unable to access CUDA repo from CUDA L4t docker container Jetson Xavier NX cuda	39	4035	November 13, 2022
apt-get update errors Jetson TX2	39	19604	October 18, 2021
Apt update failing on Ubuntu CUDA repo CUDA Setup and Installation	18	4928	July 15, 2022

"sudo apt update" not working anymore after installation of CUDA Toolkit 13.0 Update 2 Downloads

Related topics