How to install correct CUDA toolkit (8.6) for A2000 mobile on LinuxMint 22.1

Hello

I’m a bit lost in the matrix.

How to install correct CUDA toolkit (8.6) for A2000 mobile on LinuxMint 22.1

I’m easily able to install the latest via:
sudo apt install nvidia-cuda-toolkit

However this installs the latest version:

vcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Fri_Jan__6_16:45:21_PST_2023
Cuda compilation tools, release 12.0, V12.0.140
Build cuda_12.0.r12.0/compiler.32267302_0

which is not compatible with the gfx card A2000 mobile on my laptop.

So I purged the installation successfully:
sudo apt --purge remove nvidia-cuda-toolkit

According to the wikipedia matrix cuda tools 8.6 is the maximum supported for A2000 mobile:

So I tried:

sudo apt install nvidia-cuda-toolkit-8-6
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
E: Unable to locate package nvidia-cuda-toolkit-8-6

and also

sudo apt install cuda-toolkit-8-6
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
E: Unable to locate package cuda-toolkit-8-6

then I read that from this post that I could try visiting this page to build the commands.

but after running the commands:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
--2025-06-28 19:43:14--  https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
Resolving developer.download.nvidia.com (developer.download.nvidia.com)... 2.18.188.11, 2.18.188.32
Connecting to developer.download.nvidia.com (developer.download.nvidia.com)|2.18.188.11|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 4328 (4.2K) [application/x-deb]
Saving to: ‘cuda-keyring_1.1-1_all.deb’

cuda-keyring_1.1-1_all. 100%[===============================>]   4.23K  --.-KB/s    in 0s      

2025-06-28 19:43:16 (564 MB/s) - ‘cuda-keyring_1.1-1_all.deb’ saved [4328/4328]
sudo dpkg -i cuda-keyring_1.1-1_all.deb
Selecting previously unselected package cuda-keyring.
(Reading database ... 497953 files and directories currently installed.)
Preparing to unpack cuda-keyring_1.1-1_all.deb ...
Unpacking cuda-keyring (1.1-1) ...
Setting up cuda-keyring (1.1-1) ...
sudo apt-get update
Hit:1 http://dl.google.com/linux/earth/deb stable InRelease                                    
Hit:2 http://archive.ubuntu.com/ubuntu noble InRelease                                         
Hit:3 https://josm.openstreetmap.de/apt jammy InRelease                                        
Hit:4 https://brave-browser-apt-release.s3.brave.com stable InRelease                          
Hit:5 http://security.ubuntu.com/ubuntu noble-security InRelease                               
Ign:6 http://packages.linuxmint.com xia InRelease                                              
Hit:7 http://archive.ubuntu.com/ubuntu noble-updates InRelease                            
Hit:8 https://dl.google.com/linux/chrome/deb stable InRelease                             
Hit:9 http://packages.linuxmint.com xia Release                                           
Hit:10 http://archive.ubuntu.com/ubuntu noble-backports InRelease   
Get:12 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64  InRelease [1,581 B]
Get:13 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64  Packages [715 kB]
Fetched 717 kB in 5s (158 kB/s)    
Reading package lists... Done

and making sure then last command puts in the 8.6 version:

sudo apt-get -y install cuda-toolkit-8-6
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
E: Unable to locate package cuda-toolkit-8-6

I’m still lost in the matrix.

Any ideas?

Thanks

Assuming the “matrix”, you’re refering to is the section on the Wiki page titled, “GPUs supported”, I’m not sure how you come to that conclusion.

Your card is compute capability 8.6 and the architecture name is Ampere. Looking at the table I mention above, you can see 8.6 is supported from Cuda Toolkit (SDK) 11.1 through to the current version 12.9.

So you’re fine with 12.9 and it will be a while until 8.6 stops being supported.

Think of the the numbers at the beginning/end of each horizontal green row, as the min/max supported CC.

Hello rs277

Thank you for your feedback.

Actually I missed the GPUs supported table, and was just looking at the following table with Compute Capability version, and incorrectly conflated this to mean the listed GPUs were tied to a specific CUDA version (8.6)

So, OK now I see that the latest CUDA driver is backwardly compatible for this card.

My current issue is that the application I’m running in Linux, DaVinci Resolve Studio 20, is not enumerating the CUDA capability, only OpenGL.

On this dual boot system, the CUDA capability is enumerated in the application on Windows.

I’m wondering what Nvidia tools on Linux I can use to verify correct installation of CUDA?

Oh well, I tried moving to a more recent open kernel driver which has solved the issue:

sudo apt-get install nvidia-open

Unfortunately, things appear to be a bit broken now.

Although the Resolve app is running and apparently connecting to CUDA, it throws a GPU full error as soon as a video is imported.

Seems the dGPU is not seen by the system somehow.

lspci -k | grep -A 2 -i "VGA"
0000:00:02.0 VGA compatible controller: Intel Corporation TigerLake-H GT1 [UHD Graphics] (rev 01)
	DeviceName: Onboard - Video
	Subsystem: Dell TigerLake-H GT1 [UHD Graphics]
nvidia-smi
Sun Jun 29 15:10:07 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 575.57.08              Driver Version: 575.57.08      CUDA Version: 12.9     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA RTX A2000 Laptop GPU    On  |   00000000:01:00.0 Off |                  N/A |
| N/A   46C    P0            752W /   40W |      14MiB /   4096MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1251      G   /usr/lib/xorg/Xorg                        4MiB |
+-----------------------------------------------------------------------------------------+

With 575 driver installed, nvidia-prime for GPU switching is missing.

If I attempt to install nvidia-prime it appears to want to remove 575.

sudo apt install nvidia-prime
[sudo] password for tim:       
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
The following packages were automatically installed and are no longer required:
  libatomic1:i386 libbsd0:i386 libdrm-amdgpu1:i386 libdrm-intel1:i386 libdrm-radeon1:i386
  libdrm2:i386 libedit2:i386 libelf1t64:i386 libexpat1:i386 libffi8:i386 libgbm1:i386
  libglapi-mesa:i386 libicu74:i386 libllvm19:i386 libnvidia-common-575
  libnvidia-compute-575:i386 libnvidia-decode-575:i386 libnvidia-egl-gbm1
  libnvidia-egl-gbm1:i386 libnvidia-egl-wayland1 libnvidia-egl-wayland1:i386
  libnvidia-egl-xcb1 libnvidia-egl-xcb1:i386 libnvidia-egl-xlib1 libnvidia-egl-xlib1:i386
  libnvidia-encode-575 libnvidia-encode-575:i386 libnvidia-extra-575 libnvidia-fbc1-575
  libnvidia-fbc1-575:i386 libnvidia-gl-575 libnvidia-gl-575:i386 libnvidia-gpucomp-575:i386
  libpciaccess0:i386 libsensors5:i386 libstdc++6:i386 libwayland-client0:i386
  libwayland-server0:i386 libx11-6:i386 libx11-xcb1:i386 libxau6:i386 libxcb-dri2-0:i386
  libxcb-dri3-0:i386 libxcb-present0:i386 libxcb-randr0:i386 libxcb-sync1:i386
  libxcb-xfixes0:i386 libxcb1:i386 libxdmcp6:i386 libxext6:i386 libxml2:i386
  libxshmfence1:i386 mesa-libgallium:i386 nvidia-compute-utils-575 nvidia-dkms-575-open
  nvidia-firmware-575 nvidia-kernel-common-575 nvidia-kernel-source-575-open nvidia-modprobe
  nvidia-settings nvidia-utils-575 screen-resolution-extra xserver-xorg-video-nvidia-575
Use 'sudo apt autoremove' to remove them.
The following packages will be REMOVED:
  nvidia-driver-575-open
The following NEW packages will be installed:
  nvidia-prime
0 upgraded, 1 newly installed, 1 to remove and 1 not upgraded.
Need to get 10.4 kB of archives.
After this operation, 19.5 kB of additional disk space will be used.
Do you want to continue? [Y/n] n
Abort.

I do faintly recall having an issue with this once before, I had to stick on 550 in order to continue using nvidia-prime but I’m hoping there might be a solution that lets me use 575.

Sorry, I have no experience with your situation re. switching GPUs, so hopefully someone else can help.

One last comment. Although it does no harm being installed, installing the Cuda Toolkit is not required for running Resolve, the Cuda runtime is included with the driver. The toolkit is primarily for developing Cuda code.

OK it seems that the CUDA tools repo was conflicting with the runtime drivers.

First I disabled the repo in software sources:

This changed the available list of drivers in Driver Manager. After installing the latest and only available “open” driver, 570, the nvidia-prime-applet was running again, nvidia-settings displaying the expected info:

Resolve is running with CUDA processing

there are no more GPU full errors, and the video can be seen playing back in the viewport.