[GH-ISSUE #11569] AMD Instinct MI50 (gfx906) amdgpu is not supported although listed as supported #69696

Closed
opened 2026-05-04 18:52:15 -05:00 by GiteaMirror · 4 comments
Owner

Originally created by @ridvan70 on GitHub (Jul 29, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11569

What is the issue?

Is MI50 supported by Ollama?

It is listed as supported on linux in gpu.md but ollama says:

msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx1200 gfx1201 gfx908 gfx90a gfx942])" gpu_type=gfx906

Relevant log output

level=WARN source=amd_linux.go:376 msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx1200 gfx1201 gfx908 gfx90a gfx942])" gpu_type=gfx906 gpu=GP>

OS

Ubuntu 24.04.2 LTS x86_64

GPU

AMD Instinct MI50

CPU

Intel Xeon E5-2680 v3 (24) @ 3.300GHz

Ollama version

0.9.6

Originally created by @ridvan70 on GitHub (Jul 29, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11569 ### What is the issue? Is MI50 supported by Ollama? It is listed as supported on linux in gpu.md but ollama says: `msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx1200 gfx1201 gfx908 gfx90a gfx942])" gpu_type=gfx906` ### Relevant log output ```shell level=WARN source=amd_linux.go:376 msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx1200 gfx1201 gfx908 gfx90a gfx942])" gpu_type=gfx906 gpu=GP> ``` ### OS Ubuntu 24.04.2 LTS x86_64 ### GPU AMD Instinct MI50 ### CPU Intel Xeon E5-2680 v3 (24) @ 3.300GHz ### Ollama version 0.9.6
GiteaMirror added the bug label 2026-05-04 18:52:15 -05:00
Author
Owner

@rick-github commented on GitHub (Jul 29, 2025):

https://github.com/ollama/ollama/blob/main/docs/gpu.md#overrides-on-linux

<!-- gh-comment-id:3134112119 --> @rick-github commented on GitHub (Jul 29, 2025): https://github.com/ollama/ollama/blob/main/docs/gpu.md#overrides-on-linux
Author
Owner

@dhiltgen commented on GitHub (Jul 31, 2025):

@ridvan70 it looks like you have installed your own copy of ROCm 6.4 instead of using our bundled version (currently 6.3.3). AMD has removed support for this older GPU in ROCm 6.4 - https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/system-requirements.html so even though we compile support for it, at runtime, that support is being disabled on your system based on the ROCm library we discover. If you switch to an older ROCm, the GPU should continue to work. Be aware that we will be upgrading to 6.4 soon to pick up newer GPU support, so you will likely have to build from source and use an older ROCm eventually to continue having support for this GPU.

6.3.3 shows this GPU as deprecated, but still supported: https://rocm.docs.amd.com/projects/install-on-linux/en/docs-6.3.3/reference/system-requirements.html

<!-- gh-comment-id:3141418281 --> @dhiltgen commented on GitHub (Jul 31, 2025): @ridvan70 it looks like you have installed your own copy of ROCm 6.4 instead of using our [bundled version](https://github.com/ollama/ollama/releases/download/v0.10.1/ollama-linux-amd64-rocm.tgz) (currently 6.3.3). AMD has removed support for this older GPU in ROCm 6.4 - https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/system-requirements.html so even though we compile support for it, at runtime, that support is being disabled on your system based on the ROCm library we discover. If you switch to an older ROCm, the GPU should continue to work. Be aware that we will be upgrading to 6.4 soon to pick up newer GPU support, so you will likely have to build from source and use an older ROCm eventually to continue having support for this GPU. 6.3.3 shows this GPU as deprecated, but still supported: https://rocm.docs.amd.com/projects/install-on-linux/en/docs-6.3.3/reference/system-requirements.html
Author
Owner

@ridvan70 commented on GitHub (Aug 1, 2025):

@dhiltgen Yes, indeed. AMD ROCm documentation guides you to the installation of the latest version, which was the 6.4.x in my case.

Will it still be possible to use a specific version of ROCm with future releases of Ollama packaged with newer ROCm versions?

<!-- gh-comment-id:3144154401 --> @ridvan70 commented on GitHub (Aug 1, 2025): @dhiltgen Yes, indeed. AMD ROCm documentation guides you to the installation of the latest version, which was the 6.4.x in my case. Will it still be possible to use a specific version of ROCm with future releases of Ollama packaged with newer ROCm versions?
Author
Owner

@dhiltgen commented on GitHub (Aug 1, 2025):

I believe you'll have to build from source. Once we move to 6.4, my expectation is the gfx906 target will trigger an error and we'll have to remove it, so the resulting binary code wont be included.

<!-- gh-comment-id:3145174552 --> @dhiltgen commented on GitHub (Aug 1, 2025): I believe you'll have to build from source. Once we move to 6.4, my expectation is the gfx906 target will trigger an error and we'll have to remove it, so the resulting binary code wont be included.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#69696