[GH-ISSUE #7112] Linux install - better support multi-vendor systems #51030

Open
opened 2026-04-28 17:56:08 -05:00 by GiteaMirror · 3 comments
Owner

Originally created by @playday3008 on GitHub (Oct 7, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7112

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

When the script finds NVIDIA GPU software, it just exits.

defbf9425a/scripts/install.sh (L189-L192)

But in theory, a system can have multiple GPU's from multiple vendors.

In my example. I removed exit 0 from the script so it would install additional stuff for my AMD GPU.

OS

Linux

GPU

Nvidia, AMD

CPU

AMD

Ollama version

0.3.12

Originally created by @playday3008 on GitHub (Oct 7, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7112 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? When the script finds NVIDIA GPU software, it just exits. https://github.com/ollama/ollama/blob/defbf9425af8228f3420d567e9eeaa29d8ac87e3/scripts/install.sh#L189-L192 But in theory, a system can have multiple GPU's from multiple vendors. In my example. I removed `exit 0` from the script so it would install additional stuff for my AMD GPU. ### OS Linux ### GPU Nvidia, AMD ### CPU AMD ### Ollama version 0.3.12
GiteaMirror added the installfeature requestlinux labels 2026-04-28 17:56:08 -05:00
Author
Owner

@NicholasPaulick commented on GitHub (Oct 9, 2024):

Is what your asking for is install.sh to install of of the drivers for all of the GPUs detected not just stop after it checks if nvidia's are preinstalled, I don't think that install.sh does anything to specify which GPU to actually use once it's installed?

<!-- gh-comment-id:2402922203 --> @NicholasPaulick commented on GitHub (Oct 9, 2024): Is what your asking for is install.sh to install of of the drivers for all of the GPUs detected not just stop after it checks if nvidia's are preinstalled, I don't think that install.sh does anything to specify which GPU to actually use once it's installed?
Author
Owner

@dhiltgen commented on GitHub (Nov 5, 2024):

This will require a more invasive update to the install script to get right. Our device check incorrectly identifies AMD motherboards and as potentially having AMD GPUs, so simply removing the exit will result in many users downloading ROCm dependencies unnecessarily. One concept I've been considering is building upon #7262 and utilize ollama itself to do initial discovery then conditionally download the bundle(s) for the supported GPUs we find.

<!-- gh-comment-id:2458220114 --> @dhiltgen commented on GitHub (Nov 5, 2024): This will require a more invasive update to the install script to get right. Our device check incorrectly identifies AMD motherboards and as potentially having AMD GPUs, so simply removing the exit will result in many users downloading ROCm dependencies unnecessarily. One concept I've been considering is building upon #7262 and utilize ollama itself to do initial discovery then conditionally download the bundle(s) for the supported GPUs we find.
Author
Owner

@playday3008 commented on GitHub (Nov 14, 2024):

Also like for NVIDIA we're checking nvidia-smi, for AMD we can check rocminfo, at least as a clue, because there's no guarantee that it exists but if it is, there's a very high chance that we're interested in ROCm

<!-- gh-comment-id:2477224100 --> @playday3008 commented on GitHub (Nov 14, 2024): Also like for NVIDIA we're checking `nvidia-smi`, for AMD we can check `rocminfo`, at least as a clue, because there's no guarantee that it exists but if it is, there's a very high chance that we're interested in ROCm
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#51030