github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10430] Adding support for amd new GPUS 9070 and 9070 XT #32616

New Issue
Open
opened 2026-04-22 14:07:18 -05:00 by GiteaMirror · 63 comments
GiteaMirror commented 2026-04-22 14:07:18 -05:00
Owner
Copy Link

Originally created by @doomaholic on GitHub (Apr 27, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10430

Originally assigned to: @dhiltgen on GitHub.

Would like to add support for AMD new GPUS 9070 and 9070 XT, as they are both unsupported / unrecognized as gfx1201

Originally created by @doomaholic on GitHub (Apr 27, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10430 Originally assigned to: @dhiltgen on GitHub. Would like to add support for AMD new GPUS 9070 and 9070 XT, as they are both unsupported / unrecognized as gfx1201
GiteaMirror added the feature requestwindows labels 2026-04-22 14:07:26 -05:00
GiteaMirror commented 2026-04-22 14:07:31 -05:00
Author
Owner
Copy Link

@koh43 commented on GitHub (Apr 28, 2025):

I was able to run ollama with my 9070xt. Do you have the latest ROCm 6.4.0 installed?

<!-- gh-comment-id:2835675621 --> @koh43 commented on GitHub (Apr 28, 2025): I was able to run ollama with my 9070xt. Do you have the latest ROCm 6.4.0 installed?
GiteaMirror commented 2026-04-22 14:07:33 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Apr 28, 2025):

Im using 9070 non XT, and yes everything is latest edition. just to note AMD official drivers don't have ROCm enabled yet on 9070xt and 9070, The forked ollama for Amd have enabled Vulcan ROCm and enabled gfx1201 in their latest release, but i would rather it be implemented here on Main ollama

<!-- gh-comment-id:2835825036 --> @doomaholic commented on GitHub (Apr 28, 2025): Im using 9070 non XT, and yes everything is latest edition. just to note AMD official drivers don't have ROCm enabled yet on 9070xt and 9070, The forked ollama for Amd have enabled Vulcan ROCm and enabled gfx1201 in their latest release, but i would rather it be implemented here on Main ollama
GiteaMirror commented 2026-04-22 14:07:34 -05:00
Author
Owner
Copy Link

@koh43 commented on GitHub (Apr 28, 2025):

I had no issues using the main ollama, and it recognized my GPU during installation. Perhaps the forked version isn’t updated for the 9070?

<!-- gh-comment-id:2835895882 --> @koh43 commented on GitHub (Apr 28, 2025): I had no issues using the main ollama, and it recognized my GPU during installation. Perhaps the forked version isn’t updated for the 9070?
GiteaMirror commented 2026-04-22 14:07:36 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Apr 28, 2025):

hmmm .., that is very strange, well i didn't tested the fork myself, but i saw on their latest release notes they enabled " gfx1201 " that was 3 days ago which is the 9070 9070 XT support, And current ollama does not support "gfx1201" ... i don't know how you mange to make ollama recognize your gpu and use it , are you on windows ?

<!-- gh-comment-id:2836693956 --> @doomaholic commented on GitHub (Apr 28, 2025): hmmm .., that is very strange, well i didn't tested the fork myself, but i saw on their latest release notes they enabled " gfx1201 " that was 3 days ago which is the 9070 9070 XT support, And current ollama does not support "gfx1201" ... i don't know how you mange to make ollama recognize your gpu and use it , are you on windows ?
GiteaMirror commented 2026-04-22 14:07:37 -05:00
Author
Owner
Copy Link

@WhatIsAPM1989 commented on GitHub (Apr 29, 2025):

Hi! I have the same issue, same card 9070 non-xt. I have not installed ROCm separately, but it seems like it is part of the ollama, since there are some support:
Logs said msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx906])" gpu_type=gfx1201 gpu=0
Win11

<!-- gh-comment-id:2840318935 --> @WhatIsAPM1989 commented on GitHub (Apr 29, 2025): Hi! I have the same issue, same card 9070 non-xt. I have not installed ROCm separately, but it seems like it is part of the ollama, since there are some support: Logs said msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx906])" gpu_type=gfx1201 gpu=0 Win11
GiteaMirror commented 2026-04-22 14:07:40 -05:00
Author
Owner
Copy Link

@koh43 commented on GitHub (Apr 29, 2025):

@doomaholic So I'm running on Ubuntu 24.04 LTS with ROCm 6.4.0 installed. I will also try on Windows to see if that's the issue.

<!-- gh-comment-id:2840387860 --> @koh43 commented on GitHub (Apr 29, 2025): @doomaholic So I'm running on Ubuntu 24.04 LTS with ROCm 6.4.0 installed. I will also try on Windows to see if that's the issue.
GiteaMirror commented 2026-04-22 14:07:43 -05:00
Author
Owner
Copy Link

@roufpup commented on GitHub (May 4, 2025):

On linux here with a 9070XT, the service prints:
(supported types:[gfx1010 gfx1030 gfx1100 gfx1101 gfx1102 gfx900 gfx906 gfx908 gfx90a gfx942])" gpu_type=gfx1201

It would be kind of nice if we can add the 9070s as a supported card

<!-- gh-comment-id:2849035375 --> @roufpup commented on GitHub (May 4, 2025): On linux here with a 9070XT, the service prints: ` (supported types:[gfx1010 gfx1030 gfx1100 gfx1101 gfx1102 gfx900 gfx906 gfx908 gfx90a gfx942])" gpu_type=gfx1201 ` It would be kind of nice if we can add the 9070s as a supported card
GiteaMirror commented 2026-04-22 14:07:46 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (May 5, 2025):

Same on new release

127.0.0.1:11434 (version 0.6.8)
looking for compatible GPUs
unsupported Radeon iGPU detected skipping" id=0 total="18.2 GiB
amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx906])" gpu_type=gfx1201 gpu=1
no compatible GPUs were discovered

<!-- gh-comment-id:2852281017 --> @doomaholic commented on GitHub (May 5, 2025): Same on new release 127.0.0.1:11434 (version 0.6.8) looking for compatible GPUs unsupported Radeon iGPU detected skipping" id=0 total="18.2 GiB amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1151 gfx906])" gpu_type=gfx1201 gpu=1 no compatible GPUs were discovered
GiteaMirror commented 2026-04-22 14:07:48 -05:00
Author
Owner
Copy Link

@roufpup commented on GitHub (May 9, 2025):

This has been fixed for me by running ollama in a docker container with the ollama:rocm image.

<!-- gh-comment-id:2866810381 --> @roufpup commented on GitHub (May 9, 2025): This has been fixed for me by running ollama in a docker container with the ollama:rocm image.
GiteaMirror commented 2026-04-22 14:07:50 -05:00
Author
Owner
Copy Link

@brbsleepy commented on GitHub (May 10, 2025):

(I'm on windows (not WSL), and using a 9070 XT)

I just pulled the docker image for ollama:rocm, and followed the instructions here:

https://hub.docker.com/r/ollama/ollama

but I got the error
docker: Error response from daemon: error gathering device information while adding custom device "/dev/kfd": no such file or directory

I tried running docker run -d -p 11434:11434 --name ollama ollama/ollama:rocm instead and while it worked and let me pull and run llama3, it did not utilize the gpu at all.

@roufpup did you also encounter this error? or are you not on windows?

<!-- gh-comment-id:2868336361 --> @brbsleepy commented on GitHub (May 10, 2025): (I'm on windows (not WSL), and using a 9070 XT) I just pulled the docker image for ollama:rocm, and followed the instructions here: https://hub.docker.com/r/ollama/ollama but I got the error `docker: Error response from daemon: error gathering device information while adding custom device "/dev/kfd": no such file or directory` I tried running `docker run -d -p 11434:11434 --name ollama ollama/ollama:rocm` instead and while it worked and let me pull and run llama3, it did not utilize the gpu at all. @roufpup did you also encounter this error? or are you not on windows?
GiteaMirror commented 2026-04-22 14:07:52 -05:00
Author
Owner
Copy Link

@roufpup commented on GitHub (May 10, 2025):

@brb-sleepy i am on linux so i am not sure how i can help, i just made myself a custom docker compose where i attach my /dev/kfd and my /dev/dri as devices. You would have to figure out how to attach your gpu from windows to match that of a linux one probably since the docker environment expects a /dev/kfd device which on linux at least does represent your gpu.

<!-- gh-comment-id:2868337618 --> @roufpup commented on GitHub (May 10, 2025): @brb-sleepy i am on linux so i am not sure how i can help, i just made myself a custom docker compose where i attach my /dev/kfd and my /dev/dri as devices. You would have to figure out how to attach your gpu from windows to match that of a linux one probably since the docker environment expects a /dev/kfd device which on linux at least does represent your gpu.
GiteaMirror commented 2026-04-22 14:07:54 -05:00
Author
Owner
Copy Link

@koh43 commented on GitHub (May 10, 2025):

I’ve tried several setups: running Ollama on Windows, building Ollama with ROCm on Windows, and installing both Ollama and ROCm on WSL (Ubuntu 24). Unfortunately, none of them were able to access the GPU.
From what I’ve seen, the issue doesn’t seem to be on Ollama’s side, but rather with AMD—since they appear to have stopped supporting latest ROCm drivers on Windows (the latest one available is 6.2.4).
It looks like the most reliable solution for now is to use Linux for proper GPU access...😓

<!-- gh-comment-id:2869123468 --> @koh43 commented on GitHub (May 10, 2025): I’ve tried several setups: running Ollama on Windows, building Ollama with ROCm on Windows, and installing both Ollama and ROCm on WSL (Ubuntu 24). Unfortunately, none of them were able to access the GPU. From what I’ve seen, the issue doesn’t seem to be on Ollama’s side, but rather with AMD—since they appear to have stopped supporting latest ROCm drivers on Windows ([the latest one available is 6.2.4](https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html)). It looks like the most reliable solution for now is to use Linux for proper GPU access...😓
GiteaMirror commented 2026-04-22 14:07:56 -05:00
Author
Owner
Copy Link

@EvgeniySpinov commented on GitHub (May 12, 2025):

As mentioned earlier, support of 9070XT has been added from 6.4.0+ rocm. There are no binaries shared, but sources are: https://github.com/ROCm/ROCm. Perhaps if ollama recompiled with that version it would work.

<!-- gh-comment-id:2873913373 --> @EvgeniySpinov commented on GitHub (May 12, 2025): As mentioned earlier, support of 9070XT has been added from 6.4.0+ rocm. There are no binaries shared, but sources are: https://github.com/ROCm/ROCm. Perhaps if ollama recompiled with that version it would work.
GiteaMirror commented 2026-04-22 14:07:59 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Jun 3, 2025):

Will support be enabled in next release ?

<!-- gh-comment-id:2936399806 --> @doomaholic commented on GitHub (Jun 3, 2025): Will support be enabled in next release ?
GiteaMirror commented 2026-04-22 14:08:02 -05:00
Author
Owner
Copy Link

@JPetovello commented on GitHub (Jun 7, 2025):

Will support be enabled in next release ?

I would like to know this as well.

<!-- gh-comment-id:2952644073 --> @JPetovello commented on GitHub (Jun 7, 2025): > Will support be enabled in next release ? I would like to know this as well.
GiteaMirror commented 2026-04-22 14:08:05 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Jun 7, 2025):

Changelog from latest AMD driver Edition 25.6.1

AMD ROCm™ on WSL for AMD Radeon™ RX 9000 Series and AMD Radeon™ AI PRO R9700
Official support for Windows Subsystem for Linux (WSL 2) enables users with supported hardware to run workloads with AMD ROCm™ software on a Windows system, eliminating the need for dual boot set ups.
The following has been added to WSL 2:
Support for Llama.cpp
Forward Attention 2 (FA2) backward pass enablement
Support for JAX (inference)
New models: Llama 3.1, Qwen 1.5, ChatGLM 2/4
Find more information on ROCm on Radeon compatibility  here and configuration of WSL 2  here.
Installation instructions for Radeon Software with WSL 2 can be found [here]

https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-25-6-1.html

<!-- gh-comment-id:2952782512 --> @doomaholic commented on GitHub (Jun 7, 2025): Changelog from latest AMD driver Edition 25.6.1 AMD ROCm™ on WSL for AMD Radeon™ RX 9000 Series and AMD Radeon™ AI PRO R9700 Official support for Windows Subsystem for Linux (WSL 2) enables users with supported hardware to run workloads with AMD ROCm™ software on a Windows system, eliminating the need for dual boot set ups. The following has been added to WSL 2: Support for Llama.cpp Forward Attention 2 (FA2) backward pass enablement Support for JAX (inference) New models: Llama 3.1, Qwen 1.5, ChatGLM 2/4 Find more information on ROCm on Radeon compatibility  [here](https://rocm.docs.amd.com/projects/radeon/en/latest/docs/compatibility/wsl/wsl_compatibility.html) and configuration of WSL 2  [here](https://rocm.docs.amd.com/projects/radeon/en/latest/docs/limitations.html#wsl-specific-issues). Installation instructions for Radeon Software with WSL 2 can be found [here] https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-25-6-1.html
GiteaMirror commented 2026-04-22 14:08:08 -05:00
Author
Owner
Copy Link

@romain-hebert commented on GitHub (Jun 9, 2025):

Changelog from latest AMD driver Edition 25.6.1

AMD ROCm™ on WSL for AMD Radeon™ RX 9000 Series and AMD Radeon™ AI PRO R9700 Official support for Windows Subsystem for Linux (WSL 2) enables users with supported hardware to run workloads with AMD ROCm™ software on a Windows system, eliminating the need for dual boot set ups. The following has been added to WSL 2: Support for Llama.cpp Forward Attention 2 (FA2) backward pass enablement Support for JAX (inference) New models: Llama 3.1, Qwen 1.5, ChatGLM 2/4 Find more information on ROCm on Radeon compatibility  here and configuration of WSL 2  here. Installation instructions for Radeon Software with WSL 2 can be found [here]

https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-25-6-1.html

So if we want rocm on windows we just need to throw 8 gigs of ram out the window...

<!-- gh-comment-id:2957090925 --> @romain-hebert commented on GitHub (Jun 9, 2025): > Changelog from latest AMD driver Edition 25.6.1 > > AMD ROCm™ on WSL for AMD Radeon™ RX 9000 Series and AMD Radeon™ AI PRO R9700 Official support for Windows Subsystem for Linux (WSL 2) enables users with supported hardware to run workloads with AMD ROCm™ software on a Windows system, eliminating the need for dual boot set ups. The following has been added to WSL 2: Support for Llama.cpp Forward Attention 2 (FA2) backward pass enablement Support for JAX (inference) New models: Llama 3.1, Qwen 1.5, ChatGLM 2/4 Find more information on ROCm on Radeon compatibility  [here](https://rocm.docs.amd.com/projects/radeon/en/latest/docs/compatibility/wsl/wsl_compatibility.html) and configuration of WSL 2  [here](https://rocm.docs.amd.com/projects/radeon/en/latest/docs/limitations.html#wsl-specific-issues). Installation instructions for Radeon Software with WSL 2 can be found [here] > > https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-25-6-1.html So if we want rocm on windows we just need to throw 8 gigs of ram out the window...
GiteaMirror commented 2026-04-22 14:08:09 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Jun 19, 2025):

Another release ... same problem

<!-- gh-comment-id:2986677309 --> @doomaholic commented on GitHub (Jun 19, 2025): Another release ... same problem
GiteaMirror commented 2026-04-22 14:08:10 -05:00
Author
Owner
Copy Link

@YuvalPeretz commented on GitHub (Jul 14, 2025):

Following this thread.
Anyone knows to say if the 9070XT on Windows is finally supported? They also released ROCm v7 if im.not mistaken, which should support the 9070XT.

<!-- gh-comment-id:3067894785 --> @YuvalPeretz commented on GitHub (Jul 14, 2025): Following this thread. Anyone knows to say if the 9070XT on Windows is finally supported? They also released ROCm v7 if im.not mistaken, which should support the 9070XT.
GiteaMirror commented 2026-04-22 14:08:10 -05:00
Author
Owner
Copy Link

@JPetovello commented on GitHub (Jul 14, 2025):

Following this thread. Anyone knows to say if the 9070XT on Windows is finally supported? They also released ROCm v7 if im.not mistaken, which should support the 9070XT

Ollama doesn't support the RX 9000 series in Windows at this time due to the ROCm version installed with Ollama.

<!-- gh-comment-id:3070939227 --> @JPetovello commented on GitHub (Jul 14, 2025): > Following this thread. Anyone knows to say if the 9070XT on Windows is finally supported? They also released ROCm v7 if im.not mistaken, which should support the 9070XT Ollama doesn't support the RX 9000 series in Windows at this time due to the ROCm version installed with Ollama.
GiteaMirror commented 2026-04-22 14:08:12 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Jul 24, 2025):

Windows doesn't support the 9000 series via ROCm yet: the latest available version paired with ollama is 6.2.4, while 9000 series support only landed in 6.4.1.

sooo ^.-

Version Release date
6.4.2 July 21, 2025
6.4.1 May 21, 2025
6.4.0 April 11, 2025
6.3.3 February 19, 2025
6.3.2 January 28, 2025
6.3.1 December 20, 2024
6.3.0 December 3, 2024
6.2.4 November 6, 2024
<!-- gh-comment-id:3114187570 --> @doomaholic commented on GitHub (Jul 24, 2025): Windows doesn't support the 9000 series via ROCm yet: the latest available version paired with ollama is 6.2.4, while 9000 series support only landed in 6.4.1. sooo ^.- Version | Release date -- | -- 6.4.2 | July 21, 2025 6.4.1 | May 21, 2025 6.4.0 | April 11, 2025 6.3.3 | February 19, 2025 6.3.2 | January 28, 2025 6.3.1 | December 20, 2024 6.3.0 | December 3, 2024 6.2.4 | November 6, 2024
GiteaMirror commented 2026-04-22 14:08:13 -05:00
Author
Owner
Copy Link

@LSUon commented on GitHub (Jul 25, 2025):

For anyone who finds this I did manage to get ollama working on a 9070xt (Windows 11). What I did that worked for some reason: remove the main repo ollama from the computer, install the latest release of the fork at ollama-for-amd, and then replace rocblas.dll file and the rocblas/library folder with the ones found here, specifically the rocm.gfx1201.for.hip.skd.6.2.4-no-optimized.7z file. This is just following the instructions at the ollama-for-amd so I may have missed something refer there for up-to-date instructions (they seem to put installation instructions in there releases). Hope this helps someone!

<!-- gh-comment-id:3120646743 --> @LSUon commented on GitHub (Jul 25, 2025): For anyone who finds this I did manage to get ollama working on a 9070xt (Windows 11). What I did that worked for some reason: remove the main repo ollama from the computer, install the latest release of the fork at [ollama-for-amd](https://github.com/likelovewant/ollama-for-amd/releases), and then replace `rocblas.dll` file and the `rocblas/library` folder with the ones found [here](https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/tag/v0.6.2.4), specifically the `rocm.gfx1201.for.hip.skd.6.2.4-no-optimized.7z` file. This is just following the instructions at the [ollama-for-amd](https://github.com/likelovewant/ollama-for-amd) so I may have missed something refer there for up-to-date instructions (they seem to put installation instructions in there releases). Hope this helps someone!
GiteaMirror commented 2026-04-22 14:08:14 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Jul 26, 2025):

im aware of ollama-for-amd but i want Main ollama to work thats the point of this whole thread

<!-- gh-comment-id:3121264879 --> @doomaholic commented on GitHub (Jul 26, 2025): im aware of [ollama-for-amd](https://github.com/likelovewant/ollama-for-amd) but i want Main ollama to work thats the point of this whole thread
GiteaMirror commented 2026-04-22 14:08:16 -05:00
Author
Owner
Copy Link

@LSUon commented on GitHub (Jul 26, 2025):

All good it just sounded like some people weren't aware of it and this was the resource that showed up when I was trying to figure it out 👍

<!-- gh-comment-id:3121275675 --> @LSUon commented on GitHub (Jul 26, 2025): All good it just sounded like some people weren't aware of it and this was the resource that showed up when I was trying to figure it out 👍
GiteaMirror commented 2026-04-22 14:08:18 -05:00
Author
Owner
Copy Link

@EvgeniySpinov commented on GitHub (Jul 29, 2025):

For anyone who finds this I did manage to get ollama working on a 9070xt (Windows 11). What I did that worked for some reason: remove the main repo ollama from the computer, install the latest release of the fork at ollama-for-amd, and then replace rocblas.dll file and the rocblas/library folder with the ones found here, specifically the rocm.gfx1201.for.hip.skd.6.2.4-no-optimized.7z file. This is just following the instructions at the ollama-for-amd so I may have missed something refer there for up-to-date instructions (they seem to put installation instructions in there releases). Hope this helps someone!

Thank you for sharing this, I wasn't aware about the project, so given it a try.

I didn't uninstall original Ollama, just installed Ollama for AMD into different directory (using /DIR=D:..... key), replaced libraries as instructed and linked models library to newly installed Ollama.

I have 2 GPUs: Nvidia GTX 1080 and AMD Radeon 9070 XT:

time=2025-07-29T08:33:44.320-04:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-c862bc4a-1e4b-f463-f271-88f7756b32a5 library=cuda variant=v12 compute=6.1 driver=12.8 name="NVIDIA GeForce GTX 1080" total="8.0 GiB" available="7.0 GiB"
time=2025-07-29T08:33:44.320-04:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=rocm variant="" compute=gfx1201 driver=6.4 name="AMD Radeon RX 9070 XT" total="15.9 GiB" available="15.8 GiB"

Both detected successfully and I was able to run 14Gb+7Gb models without issues purely on GPUs.

Good workaround until official Ollama updates rocm.

Kudos to @LSUon

<!-- gh-comment-id:3132524613 --> @EvgeniySpinov commented on GitHub (Jul 29, 2025): > For anyone who finds this I did manage to get ollama working on a 9070xt (Windows 11). What I did that worked for some reason: remove the main repo ollama from the computer, install the latest release of the fork at [ollama-for-amd](https://github.com/likelovewant/ollama-for-amd/releases), and then replace `rocblas.dll` file and the `rocblas/library` folder with the ones found [here](https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/tag/v0.6.2.4), specifically the `rocm.gfx1201.for.hip.skd.6.2.4-no-optimized.7z` file. This is just following the instructions at the [ollama-for-amd](https://github.com/likelovewant/ollama-for-amd) so I may have missed something refer there for up-to-date instructions (they seem to put installation instructions in there releases). Hope this helps someone! Thank you for sharing this, I wasn't aware about the project, so given it a try. I didn't uninstall original Ollama, just installed Ollama for AMD into different directory (using /DIR=D:\..... key), replaced libraries as instructed and linked models library to newly installed Ollama. I have 2 GPUs: Nvidia GTX 1080 and AMD Radeon 9070 XT: ``` time=2025-07-29T08:33:44.320-04:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-c862bc4a-1e4b-f463-f271-88f7756b32a5 library=cuda variant=v12 compute=6.1 driver=12.8 name="NVIDIA GeForce GTX 1080" total="8.0 GiB" available="7.0 GiB" time=2025-07-29T08:33:44.320-04:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=rocm variant="" compute=gfx1201 driver=6.4 name="AMD Radeon RX 9070 XT" total="15.9 GiB" available="15.8 GiB" ``` Both detected successfully and I was able to run 14Gb+7Gb models without issues purely on GPUs. Good workaround until official Ollama updates rocm. Kudos to @LSUon
GiteaMirror commented 2026-04-22 14:08:19 -05:00
Author
Owner
Copy Link

@Srijan1214 commented on GitHub (Aug 7, 2025):

Would it be reasonable for this thread to be resolved once rocm support for AMD rx 9070 xt is added to windows
This GPU has been supported in Linux/ WSL since rocm 6.4.1 (https://github.com/ROCm/ROCm/releases/tag/rocm-6.4.1). But we are still waiting on windows support.

<!-- gh-comment-id:3162413721 --> @Srijan1214 commented on GitHub (Aug 7, 2025): Would it be reasonable for this thread to be resolved once rocm support for AMD rx 9070 xt is added to windows This GPU has been supported in Linux/ WSL since rocm 6.4.1 (https://github.com/ROCm/ROCm/releases/tag/rocm-6.4.1). But we are still waiting on windows support.
GiteaMirror commented 2026-04-22 14:08:20 -05:00
Author
Owner
Copy Link

@tinyboxvk commented on GitHub (Aug 18, 2025):

AMD HIP SDK for Windows 6.4.2 has been released. https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html

<!-- gh-comment-id:3198710136 --> @tinyboxvk commented on GitHub (Aug 18, 2025): AMD HIP SDK for Windows 6.4.2 has been released. https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html
GiteaMirror commented 2026-04-22 14:08:22 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Aug 21, 2025):

@dhiltgen please merge <3

<!-- gh-comment-id:3212239272 --> @doomaholic commented on GitHub (Aug 21, 2025): @dhiltgen please merge <3
GiteaMirror commented 2026-04-22 14:08:22 -05:00
Author
Owner
Copy Link

@leacar21 commented on GitHub (Aug 23, 2025):

Is there a PR to resolve this? If not, does anyone know which file needs to be modified?

<!-- gh-comment-id:3216022122 --> @leacar21 commented on GitHub (Aug 23, 2025): Is there a PR to resolve this? If not, does anyone know which file needs to be modified?
GiteaMirror commented 2026-04-22 14:08:24 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Sep 5, 2025):

Hello ?

<!-- gh-comment-id:3256817442 --> @doomaholic commented on GitHub (Sep 5, 2025): Hello ?
GiteaMirror commented 2026-04-22 14:08:26 -05:00
Author
Owner
Copy Link

@jclsn commented on GitHub (Sep 6, 2025):

For me the GPU is recognized, but there is an error message /sys/module/amdgpu/version no such file or directory. The models then also run on the CPU.

Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.172+02:00 level=INFO source=images.go:484 msg="total unused blobs removed: 0"
Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.173+02:00 level=INFO source=routes.go:1384 msg="Listening on [::]:11434 (version 0.11.8)"
Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.173+02:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs"
Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=WARN source=amd_linux.go:61 msg="ollama recommends running the https://www.amd.com/en/support/download/linux-drivers.html" error="amdgpu version file missing: /sys/module/amdgpu/version stat /sys/module/amdgpu/version: no such file or directory"
Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=INFO source=amd_linux.go:392 msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=12.0.1
Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-35bbdf7546b4136c library=rocm variant="" compute=gfx1201 driver=0.0 name=1002:7550 total="15.9 GiB" available="13.6 GiB"
Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=INFO source=routes.go:1425 msg="entering low vram mode" "total vram"="15.9 GiB" threshold="20.0 GiB"
<!-- gh-comment-id:3263267870 --> @jclsn commented on GitHub (Sep 6, 2025): For me the GPU is recognized, but there is an error message `/sys/module/amdgpu/version no such file or directory`. The models then also run on the CPU. ``` Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.172+02:00 level=INFO source=images.go:484 msg="total unused blobs removed: 0" Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.173+02:00 level=INFO source=routes.go:1384 msg="Listening on [::]:11434 (version 0.11.8)" Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.173+02:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs" Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=WARN source=amd_linux.go:61 msg="ollama recommends running the https://www.amd.com/en/support/download/linux-drivers.html" error="amdgpu version file missing: /sys/module/amdgpu/version stat /sys/module/amdgpu/version: no such file or directory" Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=INFO source=amd_linux.go:392 msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=12.0.1 Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=INFO source=types.go:130 msg="inference compute" id=GPU-35bbdf7546b4136c library=rocm variant="" compute=gfx1201 driver=0.0 name=1002:7550 total="15.9 GiB" available="13.6 GiB" Sep 07 00:48:41 precision5810 ollama[105314]: time=2025-09-07T00:48:41.218+02:00 level=INFO source=routes.go:1425 msg="entering low vram mode" "total vram"="15.9 GiB" threshold="20.0 GiB" ```
GiteaMirror commented 2026-04-22 14:08:28 -05:00
Author
Owner
Copy Link

@leacar21 commented on GitHub (Sep 12, 2025):

Hello @dhiltgen. How are you? Do you think I can help in any way with this issue?

<!-- gh-comment-id:3283348884 --> @leacar21 commented on GitHub (Sep 12, 2025): Hello @dhiltgen. How are you? Do you think I can help in any way with this issue?
GiteaMirror commented 2026-04-22 14:08:29 -05:00
Author
Owner
Copy Link

@martintw123 commented on GitHub (Sep 15, 2025):

Would be nice to have this on main instead of doing it funky via https://github.com/likelovewant/ollama-for-amd

<!-- gh-comment-id:3291348119 --> @martintw123 commented on GitHub (Sep 15, 2025): Would be nice to have this on main instead of doing it funky via https://github.com/likelovewant/ollama-for-amd
GiteaMirror commented 2026-04-22 14:08:31 -05:00
Author
Owner
Copy Link

@xzxshmuner-boop commented on GitHub (Sep 16, 2025):

Currently, the ROCm version for Windows has been updated to 6.4.2, which supports AMD's 90 series graphics cards. It is recommended that the official version of Ollama also follow suit and provide support.

<!-- gh-comment-id:3299157569 --> @xzxshmuner-boop commented on GitHub (Sep 16, 2025): Currently, the ROCm version for Windows has been updated to 6.4.2, which supports AMD's 90 series graphics cards. It is recommended that the official version of Ollama also follow suit and provide support.
GiteaMirror commented 2026-04-22 14:08:34 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Oct 3, 2025):

this has to be a joke right

<!-- gh-comment-id:3367514814 --> @doomaholic commented on GitHub (Oct 3, 2025): this has to be a joke right
GiteaMirror commented 2026-04-22 14:08:36 -05:00
Author
Owner
Copy Link

@JPetovello commented on GitHub (Oct 22, 2025):

Hello? When are we going to see proper support for the AMD R9000 series?

<!-- gh-comment-id:3431456638 --> @JPetovello commented on GitHub (Oct 22, 2025): Hello? When are we going to see proper support for the AMD R9000 series?
GiteaMirror commented 2026-04-22 14:08:38 -05:00
Author
Owner
Copy Link

@jclsn commented on GitHub (Oct 22, 2025):

It actually works quite well on Linux now. I was able to configure it. Can't recommend this GPU for LLMs though. It uses 70W when a model is idling in VRAM and also makes some cricket-like sound when processing a model.

This is a model just idling in VRAM:

Image

This is all that was required for me

❯ sudo systemctl cat ollama
# /usr/lib/systemd/system/ollama.service
[Unit]
Description=Ollama Service
Wants=network-online.target
After=network.target network-online.target

[Service]
ExecStart=/usr/bin/ollama serve
WorkingDirectory=/var/lib/ollama
Environment="HOME=/var/lib/ollama"
Environment="OLLAMA_MODELS=/var/lib/ollama"
User=ollama
Group=ollama
Restart=on-failure
RestartSec=3
RestartPreventExitStatus=1
Type=simple
PrivateTmp=yes
ProtectSystem=full
ProtectHome=yes

[Install]
WantedBy=multi-user.target

# /etc/systemd/system/ollama.service.d/override.conf
[Service]
Environment="HSA_OVERRIDE_GFX_VERSION=12.0.1"

and the packages required on Arch Linux

❯ pacman -Qs rocm
local/hipblas 6.4.4-1
    ROCm BLAS marshalling library
local/hsa-rocr 6.4.4-1
    HSA Runtime API and runtime for ROCm
local/ollama-rocm 0.12.6-1
    Create, run and share large language models (LLMs) with ROCm
local/rocblas 6.4.4-1
    Next generation BLAS implementation for ROCm platform
local/rocm-core 6.4.4-1
    AMD ROCm core package (version files)
local/rocm-device-libs 2:6.4.4-1
    AMD specific device-side language runtime libraries
local/rocm-llvm 2:6.4.4-1
    Radeon Open Compute - LLVM toolchain (llvm, clang, lld)
local/rocm-smi-lib 6.4.4-1
    ROCm System Management Interface Library
local/rocminfo 6.4.4-1
    ROCm Application for Reporting System Info
local/rocsolver 6.4.4-1
    Subset of LAPACK functionality on the ROCm platform
local/rocsparse 6.4.4-2
    BLAS for sparse computation on top of ROCm
local/roctracer 6.4.4-1
    ROCm tracer library for performance tracing
<!-- gh-comment-id:3431475225 --> @jclsn commented on GitHub (Oct 22, 2025): It actually works quite well on Linux now. I was able to configure it. Can't recommend this GPU for LLMs though. It uses 70W when a model is idling in VRAM and also makes some cricket-like sound when processing a model. This is a model just idling in VRAM: <img width="790" height="309" alt="Image" src="https://github.com/user-attachments/assets/107326ec-5f93-497a-8d6d-203db123749d" /> This is all that was required for me ```systemd ❯ sudo systemctl cat ollama # /usr/lib/systemd/system/ollama.service [Unit] Description=Ollama Service Wants=network-online.target After=network.target network-online.target [Service] ExecStart=/usr/bin/ollama serve WorkingDirectory=/var/lib/ollama Environment="HOME=/var/lib/ollama" Environment="OLLAMA_MODELS=/var/lib/ollama" User=ollama Group=ollama Restart=on-failure RestartSec=3 RestartPreventExitStatus=1 Type=simple PrivateTmp=yes ProtectSystem=full ProtectHome=yes [Install] WantedBy=multi-user.target # /etc/systemd/system/ollama.service.d/override.conf [Service] Environment="HSA_OVERRIDE_GFX_VERSION=12.0.1" ``` and the packages required on Arch Linux ``` ❯ pacman -Qs rocm local/hipblas 6.4.4-1 ROCm BLAS marshalling library local/hsa-rocr 6.4.4-1 HSA Runtime API and runtime for ROCm local/ollama-rocm 0.12.6-1 Create, run and share large language models (LLMs) with ROCm local/rocblas 6.4.4-1 Next generation BLAS implementation for ROCm platform local/rocm-core 6.4.4-1 AMD ROCm core package (version files) local/rocm-device-libs 2:6.4.4-1 AMD specific device-side language runtime libraries local/rocm-llvm 2:6.4.4-1 Radeon Open Compute - LLVM toolchain (llvm, clang, lld) local/rocm-smi-lib 6.4.4-1 ROCm System Management Interface Library local/rocminfo 6.4.4-1 ROCm Application for Reporting System Info local/rocsolver 6.4.4-1 Subset of LAPACK functionality on the ROCm platform local/rocsparse 6.4.4-2 BLAS for sparse computation on top of ROCm local/roctracer 6.4.4-1 ROCm tracer library for performance tracing ```
GiteaMirror commented 2026-04-22 14:08:38 -05:00
Author
Owner
Copy Link

@EvgeniySpinov commented on GitHub (Oct 22, 2025):

I'm using this GPU for ollama-amd and for llama.cpp (Windows 10) and there is defintely no issues with the GPU. It's fast (when model fits into VRAM), quiet (comsumes around 100W) and doesn't produce any sounds. This is when using rocm backend. When using Vulkan (I have 2 GPUs: NVidia and AMD) it consumes less power and slower, but that backend is a different story, specific to llama.cpp.

So, we need 9000 series support for Windows - it works well on other solutions.

<!-- gh-comment-id:3432535761 --> @EvgeniySpinov commented on GitHub (Oct 22, 2025): I'm using this GPU for ollama-amd and for llama.cpp (Windows 10) and there is defintely no issues with the GPU. It's fast (when model fits into VRAM), quiet (comsumes around 100W) and doesn't produce any sounds. This is when using rocm backend. When using Vulkan (I have 2 GPUs: NVidia and AMD) it consumes less power and slower, but that backend is a different story, specific to llama.cpp. So, we need 9000 series support for Windows - it works well on other solutions.
GiteaMirror commented 2026-04-22 14:08:40 -05:00
Author
Owner
Copy Link

@romain-hebert commented on GitHub (Oct 25, 2025):

I'm using this GPU for ollama-amd and for llama.cpp (Windows 10) and there is defintely no issues with the GPU. It's fast (when model fits into VRAM), quiet (comsumes around 100W) and doesn't produce any sounds. This is when using rocm backend. When using Vulkan (I have 2 GPUs: NVidia and AMD) it consumes less power and slower, but that backend is a different story, specific to llama.cpp.

So, we need 9000 series support for Windows - it works well on other solutions.

When using ollama-for-amd, does it also use up a lot of system RAM for you ?
I tried it and it was unusable for me because of the RAM usage, that didn’t happen when I had an Nvidia gpu on standard ollama

<!-- gh-comment-id:3446559987 --> @romain-hebert commented on GitHub (Oct 25, 2025): > I'm using this GPU for ollama-amd and for llama.cpp (Windows 10) and there is defintely no issues with the GPU. It's fast (when model fits into VRAM), quiet (comsumes around 100W) and doesn't produce any sounds. This is when using rocm backend. When using Vulkan (I have 2 GPUs: NVidia and AMD) it consumes less power and slower, but that backend is a different story, specific to llama.cpp. > > So, we need 9000 series support for Windows - it works well on other solutions. When using ollama-for-amd, does it also use up a lot of system RAM for you ? I tried it and it was unusable for me because of the RAM usage, that didn’t happen when I had an Nvidia gpu on standard ollama
GiteaMirror commented 2026-04-22 14:08:41 -05:00
Author
Owner
Copy Link

@EvgeniySpinov commented on GitHub (Oct 25, 2025):

No, I do not observe such behavior.

Image Image
<!-- gh-comment-id:3446882806 --> @EvgeniySpinov commented on GitHub (Oct 25, 2025): No, I do not observe such behavior. <img width="1364" height="98" alt="Image" src="https://github.com/user-attachments/assets/3b8799d9-4403-4c8c-a520-a32010aa75a6" /> <img width="945" height="45" alt="Image" src="https://github.com/user-attachments/assets/c466d0ef-f3de-4d33-a589-c19e39707d7a" />
GiteaMirror commented 2026-04-22 14:08:42 -05:00
Author
Owner
Copy Link

@r3tr0g4m3r commented on GitHub (Oct 28, 2025):

To make it use less resources try to add this environment variable to ollama:

Environment="GPU_MAX_HW_QUEUES=1"

It actually works quite well on Linux now. I was able to configure it. Can't recommend this GPU for LLMs though. It uses 70W when a model is idling in VRAM and also makes some cricket-like sound when processing a model.

<!-- gh-comment-id:3456372794 --> @r3tr0g4m3r commented on GitHub (Oct 28, 2025): To make it use less resources try to add this environment variable to ollama: `Environment="GPU_MAX_HW_QUEUES=1"` > It actually works quite well on Linux now. I was able to configure it. Can't recommend this GPU for LLMs though. It uses 70W when a model is idling in VRAM and also makes some cricket-like sound when processing a model.
GiteaMirror commented 2026-04-22 14:08:44 -05:00
Author
Owner
Copy Link

@jclsn commented on GitHub (Oct 28, 2025):

To make it use less resources try to add this environment variable to ollama:

Environment="GPU_MAX_HW_QUEUES=1"

It actually works quite well on Linux now. I was able to configure it. Can't recommend this GPU for LLMs though. It uses 70W when a model is idling in VRAM and also makes some cricket-like sound when processing a model.

Thanks, this actually makes it run much cooler and consume only 20W when idling a model, which isn't much more than without the model loaded.

<!-- gh-comment-id:3458196589 --> @jclsn commented on GitHub (Oct 28, 2025): > To make it use less resources try to add this environment variable to ollama: > > `Environment="GPU_MAX_HW_QUEUES=1"` > > > It actually works quite well on Linux now. I was able to configure it. Can't recommend this GPU for LLMs though. It uses 70W when a model is idling in VRAM and also makes some cricket-like sound when processing a model. Thanks, this actually makes it run much cooler and consume only 20W when idling a model, which isn't much more than without the model loaded.
GiteaMirror commented 2026-04-22 14:08:47 -05:00
Author
Owner
Copy Link

@JPetovello commented on GitHub (Oct 28, 2025):

And we still wait...

<!-- gh-comment-id:3458400817 --> @JPetovello commented on GitHub (Oct 28, 2025): And we still wait...
GiteaMirror commented 2026-04-22 14:08:49 -05:00
Author
Owner
Copy Link

@doomaholic commented on GitHub (Oct 29, 2025):

bro its working on LM studio already

<!-- gh-comment-id:3461284314 --> @doomaholic commented on GitHub (Oct 29, 2025): bro its working on LM studio already
GiteaMirror commented 2026-04-22 14:08:51 -05:00
Author
Owner
Copy Link

@KineticLogic commented on GitHub (Dec 12, 2025):

Oh, good, so I'm not alone! I am also having trouble with Ollama detecting and using my 9070XT. Luckily, no other issues, but it would be nice to utilize the full power without fiddling with forks. :)

<!-- gh-comment-id:3644712438 --> @KineticLogic commented on GitHub (Dec 12, 2025): Oh, good, so I'm not alone! I am also having trouble with Ollama detecting and using my 9070XT. Luckily, no other issues, but it would be nice to utilize the full power without fiddling with forks. :)
GiteaMirror commented 2026-04-22 14:08:53 -05:00
Author
Owner
Copy Link

@Cresius34 commented on GitHub (Dec 19, 2025):

You can push ollama to use Vulkan instead rocm for the moment..

<!-- gh-comment-id:3676981150 --> @Cresius34 commented on GitHub (Dec 19, 2025): You can push ollama to use Vulkan instead rocm for the moment..
GiteaMirror commented 2026-04-22 14:08:54 -05:00
Author
Owner
Copy Link

@rhirne12 commented on GitHub (Dec 30, 2025):

So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though.

I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK).

Ollama now correctly finds my GPU.

<!-- gh-comment-id:3700027521 --> @rhirne12 commented on GitHub (Dec 30, 2025): So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though. I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK). Ollama now correctly finds my GPU.
GiteaMirror commented 2026-04-22 14:08:57 -05:00
Author
Owner
Copy Link

@Cresius34 commented on GitHub (Dec 30, 2025):

So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though.

I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK).

Ollama now correctly finds my GPU.

Interesting, but does't work whit my 9700 AI Pro

<!-- gh-comment-id:3700119850 --> @Cresius34 commented on GitHub (Dec 30, 2025): > So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though. > > I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK). > > Ollama now correctly finds my GPU. Interesting, but does't work whit my 9700 AI Pro
GiteaMirror commented 2026-04-22 14:09:00 -05:00
Author
Owner
Copy Link

@Cresius34 commented on GitHub (Dec 30, 2025):

okay, i understand nothing, but after replacing rocm folder in ollama directory (that have broken rocm in ollama).
I have re-installed ollama, I confirm it work fine !

So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though.

I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK).

Ollama now correctly finds my GPU.

Image
<!-- gh-comment-id:3700251149 --> @Cresius34 commented on GitHub (Dec 30, 2025): okay, i understand nothing, but after replacing rocm folder in ollama directory (that have broken rocm in ollama). I have re-installed ollama, I confirm it work fine ! > So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though. > > I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK). > > Ollama now correctly finds my GPU. <img width="3900" height="2066" alt="Image" src="https://github.com/user-attachments/assets/877d9a7a-4e25-45b2-9b24-3e46e99bd84a" />
GiteaMirror commented 2026-04-22 14:09:02 -05:00
Author
Owner
Copy Link

@trtr6842-git commented on GitHub (Jan 3, 2026):

On windows 11 with an RX9070 ollama used to use GPU and models ran just fine. Suddenly ollama is using only CPU. I've tried the instructions and rocm files here: https://github.com/likelovewant/ollama-for-amd/releases/tag/v0.6.3 and no luck.

I installed AMD HIP SDK (6.4.2), and tried those rocm files, and adding the "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" environment variable with no luck.

running ollama serve shows that ollama is not getting any string for the gpu_type, gpu0 is the RX9070. Any ideas?

C:\Users\ttyle>ollama serve
2026/01/03 11:16:30 routes.go:1230: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:2048 OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:C:\\LLaModels OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]"
time=2026-01-03T11:16:30.687-08:00 level=INFO source=images.go:433 msg="total blobs: 41"
time=2026-01-03T11:16:30.688-08:00 level=INFO source=images.go:440 msg="total unused blobs removed: 0"
time=2026-01-03T11:16:30.688-08:00 level=INFO source=routes.go:1297 msg="Listening on 127.0.0.1:11434 (version 0.6.3)"
time=2026-01-03T11:16:30.689-08:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs"
time=2026-01-03T11:16:30.689-08:00 level=INFO source=gpu_windows.go:167 msg=packages count=1
time=2026-01-03T11:16:30.689-08:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=12 efficiency=0 threads=24
time=2026-01-03T11:16:30.843-08:00 level=WARN source=amd_windows.go:138 msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1150 gfx1151 gfx1200 gfx1201 gfx906])" gpu_type="" gpu=0 library=C:\Users\ttyle\AppData\Local\Programs\Ollama\lib\ollama\rocm
time=2026-01-03T11:16:30.939-08:00 level=WARN source=amd_windows.go:138 msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1150 gfx1151 gfx1200 gfx1201 gfx906])" gpu_type="\x01" gpu=1 library=C:\Users\ttyle\AppData\Local\Programs\Ollama\lib\ollama\rocm
time=2026-01-03T11:16:30.940-08:00 level=INFO source=gpu.go:377 msg="no compatible GPUs were discovered"
time=2026-01-03T11:16:30.940-08:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=cpu variant="" compute="" driver=0.0 name="" total="125.6 GiB" available="111.2 GiB"
<!-- gh-comment-id:3707294717 --> @trtr6842-git commented on GitHub (Jan 3, 2026): On windows 11 with an RX9070 ollama used to use GPU and models ran just fine. Suddenly ollama is using only CPU. I've tried the instructions and rocm files here: [https://github.com/likelovewant/ollama-for-amd/releases/tag/v0.6.3](url) and no luck. I installed AMD HIP SDK (6.4.2), and tried those rocm files, and adding the `"ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library"` environment variable with no luck. running ollama serve shows that ollama is not getting any string for the gpu_type, gpu0 is the RX9070. Any ideas? ``` C:\Users\ttyle>ollama serve 2026/01/03 11:16:30 routes.go:1230: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:2048 OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:C:\\LLaModels OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:]" time=2026-01-03T11:16:30.687-08:00 level=INFO source=images.go:433 msg="total blobs: 41" time=2026-01-03T11:16:30.688-08:00 level=INFO source=images.go:440 msg="total unused blobs removed: 0" time=2026-01-03T11:16:30.688-08:00 level=INFO source=routes.go:1297 msg="Listening on 127.0.0.1:11434 (version 0.6.3)" time=2026-01-03T11:16:30.689-08:00 level=INFO source=gpu.go:217 msg="looking for compatible GPUs" time=2026-01-03T11:16:30.689-08:00 level=INFO source=gpu_windows.go:167 msg=packages count=1 time=2026-01-03T11:16:30.689-08:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=12 efficiency=0 threads=24 time=2026-01-03T11:16:30.843-08:00 level=WARN source=amd_windows.go:138 msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1150 gfx1151 gfx1200 gfx1201 gfx906])" gpu_type="" gpu=0 library=C:\Users\ttyle\AppData\Local\Programs\Ollama\lib\ollama\rocm time=2026-01-03T11:16:30.939-08:00 level=WARN source=amd_windows.go:138 msg="amdgpu is not supported (supported types:[gfx1030 gfx1100 gfx1101 gfx1102 gfx1150 gfx1151 gfx1200 gfx1201 gfx906])" gpu_type="\x01" gpu=1 library=C:\Users\ttyle\AppData\Local\Programs\Ollama\lib\ollama\rocm time=2026-01-03T11:16:30.940-08:00 level=INFO source=gpu.go:377 msg="no compatible GPUs were discovered" time=2026-01-03T11:16:30.940-08:00 level=INFO source=types.go:130 msg="inference compute" id=0 library=cpu variant="" compute="" driver=0.0 name="" total="125.6 GiB" available="111.2 GiB" ```
GiteaMirror commented 2026-04-22 14:09:03 -05:00
Author
Owner
Copy Link

@Cresius34 commented on GitHub (Jan 3, 2026):

For that i see, ollama try to use that own rocm library, have you put the Path in user or system ?

Image

And after that, try to reinstall ollama on the already installed, that has works for me ^^'

<!-- gh-comment-id:3707303014 --> @Cresius34 commented on GitHub (Jan 3, 2026): For that i see, ollama try to use that own rocm library, have you put the Path in user or system ? <img width="1118" height="920" alt="Image" src="https://github.com/user-attachments/assets/7466bc1d-2358-4890-a32f-3cb0b881b2dd" /> And after that, try to reinstall ollama on the already installed, that has works for me ^^'
GiteaMirror commented 2026-04-22 14:09:07 -05:00
Author
Owner
Copy Link

@trtr6842-git commented on GitHub (Jan 3, 2026):

That worked, thank you!
To re-cap what ended up working

  1. Installed AMD HIP SDK (6.4.2)
  2. Added a new user environment variable ROCBLAS_TENSILE_LIBPATH pointing to C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library
  3. uninstalled my current version of ollama
  4. installed the latest version of ollama (0.13.5), NOT the ollama for AMD version.
  5. Models worked and ran on the GPU as expected!
<!-- gh-comment-id:3707329597 --> @trtr6842-git commented on GitHub (Jan 3, 2026): That worked, thank you! To re-cap what ended up working 1. Installed AMD HIP SDK (6.4.2) 2. Added a new user environment variable `ROCBLAS_TENSILE_LIBPATH` pointing to `C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library` 3. uninstalled my current version of ollama 4. installed the latest version of ollama (0.13.5), NOT the ollama for AMD version. 5. Models worked and ran on the GPU as expected!
GiteaMirror commented 2026-04-22 14:09:09 -05:00
Author
Owner
Copy Link

@Cresius34 commented on GitHub (Jan 3, 2026):

Great ! Have fun :)
I think, ollama have an first launch configuration file and need to rewrite them.

<!-- gh-comment-id:3707332716 --> @Cresius34 commented on GitHub (Jan 3, 2026): Great ! Have fun :) I think, ollama have an first launch configuration file and need to rewrite them.
GiteaMirror commented 2026-04-22 14:09:10 -05:00
Author
Owner
Copy Link

@YuvalPeretz commented on GitHub (Jan 20, 2026):

@trtr6842-git Even after following you instructions directly, it still runs it for me on the CPU.

<!-- gh-comment-id:3771582836 --> @YuvalPeretz commented on GitHub (Jan 20, 2026): @trtr6842-git Even after following you instructions directly, it still runs it for me on the CPU.
GiteaMirror commented 2026-04-22 14:09:11 -05:00
Author
Owner
Copy Link

@twaaaadahardeep commented on GitHub (Jan 25, 2026):

So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though.

I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK).

Ollama now correctly finds my GPU.

OMG thanks, after fooling around for a couple of days trying to get my rx 9060xt to work with ollama, this made it work finally. Big thanks. It was so easy. 🙏

<!-- gh-comment-id:3797091649 --> @twaaaadahardeep commented on GitHub (Jan 25, 2026): > So I use a 9060xt on Windows and I was also having issues with it detecting my card. I discovered that Ollama was detecting the GPU, but couldn't load the correct tensor file (it's actually missing the tensor files). I was able to get it to detect though. > > I installed the latest Windows AMD HIP SDK (6.4.2) which supports the 9060 and 9070 cards. Then in my environment variables, I set "ROCBLAS_TENSILE_LIBPATH" to "C:\Program Files\AMD\ROCm\6.4\bin\rocblas\library" (default HIP SDK install directory - set to wherever you installed HIP SDK). > > Ollama now correctly finds my GPU. OMG thanks, after fooling around for a couple of days trying to get my rx 9060xt to work with ollama, this made it work finally. Big thanks. It was so easy. 🙏
GiteaMirror commented 2026-04-22 14:09:12 -05:00
Author
Owner
Copy Link

@solokiran commented on GitHub (Feb 3, 2026):

Hi,
I just want to add my issue.

When ever I chat, my computer keeps on crashing ( display off and computer freeze).

My GPU is RX 9070 XT and CPU is i7 14700, RAM is 64 GB. Windows 11.

I have installed the latest graphic driver (Adrenaline 26.1.1) included the AI bundle from the installation.

Installed ollama (0.15.2) and created ROCBLAS_TENSILE_LIBPATH env variable and assigned the value "AppData/Local/AMD/AI_Bundle/ComfyUI/venv/Lib/site-packages/_rocm_sdk_libraries_custom/bin/rocblas/library"

I am loading the model llama3.1:8b
Model is getting loaded to GPU and I am able to chat.

Please note: chatting from command line is working great.

When I integrate it to VS Code via continue and chat. The computer crashes.

Few ChatGPT and Copilot answers says the GPU is power hunger and 750 W PSU may not be enough.
But I ran Hogwarts legacy game in 4K without any issue.

Few answers says to reduce the layers count. I limited to 20. Still crashing.

No matter what I do, crash is end result.

Edit:
Not just while using VS code.
Even If I use LM studio or Goose agent, it will crash.

<!-- gh-comment-id:3841147101 --> @solokiran commented on GitHub (Feb 3, 2026): Hi, I just want to add my issue. When ever I chat, my computer keeps on **crashing** ( display off and computer freeze). My GPU is RX 9070 XT and CPU is i7 14700, RAM is 64 GB. Windows 11. I have installed the latest graphic driver (**Adrenaline 26.1.1**) included the AI bundle from the installation. Installed ollama **(0.15.2)** and created ROCBLAS_TENSILE_LIBPATH env variable and assigned the value "AppData/Local/AMD/AI_Bundle/ComfyUI/venv/Lib/site-packages/_rocm_sdk_libraries_custom/bin/rocblas/library" I am loading the model **llama3.1:8b** Model is getting loaded to GPU and I am able to chat. Please note: chatting from command line is working great. When I integrate it to VS Code via continue and chat. The computer crashes. Few ChatGPT and Copilot answers says the GPU is power hunger and 750 W PSU may not be enough. But I ran Hogwarts legacy game in 4K without any issue. Few answers says to reduce the layers count. I limited to 20. Still crashing. No matter what I do, crash is end result. Edit: Not just while using VS code. Even If I use LM studio or Goose agent, it will crash.
GiteaMirror commented 2026-04-22 14:09:14 -05:00
Author
Owner
Copy Link

@Cresius34 commented on GitHub (Feb 3, 2026):

Whether or not it crashes depends on the nature of the "crash". If it’s a power supply issue, your computer would shut down with a black screen. 750W is more than enough for your system, which likely consumes between 450 and 500W under full CPU/GPU load.

I suspect the integration with VS Code is continuously loading memory until it’s full. In this case, I’d recommend keeping an eye on memory usage and looking more into drivers/software issues.

<!-- gh-comment-id:3841213794 --> @Cresius34 commented on GitHub (Feb 3, 2026): Whether or not it crashes depends on the nature of the "crash". If it’s a power supply issue, your computer would shut down with a black screen. 750W is more than enough for your system, which likely consumes between 450 and 500W under full CPU/GPU load. I suspect the integration with VS Code is continuously loading memory until it’s full. In this case, I’d recommend keeping an eye on memory usage and looking more into drivers/software issues.
GiteaMirror commented 2026-04-22 14:09:16 -05:00
Author
Owner
Copy Link

@androiddrew commented on GitHub (Feb 5, 2026):

@doomaholic checkout https://github.com/ollama/ollama/issues/12908#issuecomment-3854823325 using ROCm 7.2 on the host. The RDNA4 cards aren't going to have any fine tuned kernels though. I have been raising this to anyone I can on the AMD discord.

<!-- gh-comment-id:3856888051 --> @androiddrew commented on GitHub (Feb 5, 2026): @doomaholic checkout https://github.com/ollama/ollama/issues/12908#issuecomment-3854823325 using ROCm 7.2 on the host. The RDNA4 cards aren't going to have any fine tuned kernels though. I have been raising this to anyone I can on the AMD discord.
GiteaMirror commented 2026-04-22 14:09:18 -05:00
Author
Owner
Copy Link

@prabhdatnoor commented on GitHub (Mar 7, 2026):

Hey everyone,

I was trying to run Ollama on WSL Ubuntu and was not able to get it to use the GPU (I have 9070XT).

I think it may be because the ROCM with Ollama is not the latest one, so I compiled with the newer ROCM, and it is working now!

1. Prerequisites:

2. Clone Ollama:

git clone https://github.com/ollama/ollama

cd ollama

3. Build GPU backend:

Erase any prev builds

rm -rf build

cmake --preset "ROCm 6" \
    -DAMDGPU_TARGETS="gfx1201" \
    -DGPU_TARGETS="gfx1201" \
    -DCMAKE_PREFIX_PATH=/opt/rocm-7.2.0 \
    -Dhip_DIR=/opt/rocm-7.2.0/lib/cmake/hip

cmake --build build

4. Build Ollama binary:

go clean -cache

go build -o ollama .

5. Run

./ollama serve

Since Ollama wasn't working, I was trying to use llama.cpp but then I had to compile that for my machine, but then I thought why not just compile ollama!

Flags:

  • --preset "ROCm 6" - loads Ollama's predefined ROCm build configuration
  • -DAMDGPU_TARGETS / -DGPU_TARGETS - specifies your GPU architecture (both needed for ROCm compatibility)
  • -DCMAKE_PREFIX_PATH - tells CMake where to find your ROCm installation
  • -Dhip_DIR - points directly to the HIP cmake config file

There may be a better way to do it, but for now this worked for me, and hopefully this could be helpful to others that were stuck!

<!-- gh-comment-id:4017145521 --> @prabhdatnoor commented on GitHub (Mar 7, 2026): Hey everyone, I was trying to run Ollama on WSL Ubuntu and was not able to get it to use the GPU (I have 9070XT). I think it may be because the ROCM with Ollama is not the latest one, so I compiled with the newer ROCM, and it is working now! #### 1. Prerequisites: - All listed on [docs/development.md](https://github.com/ollama/ollama/blob/main/docs/development.md) - [ROCM on WSL](https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/wsl/install-radeon.html) #### 2. Clone Ollama: ``` git clone https://github.com/ollama/ollama ``` ``` cd ollama ``` #### 3. Build GPU backend: Erase any prev builds ``` rm -rf build ``` ``` cmake --preset "ROCm 6" \ -DAMDGPU_TARGETS="gfx1201" \ -DGPU_TARGETS="gfx1201" \ -DCMAKE_PREFIX_PATH=/opt/rocm-7.2.0 \ -Dhip_DIR=/opt/rocm-7.2.0/lib/cmake/hip ``` ``` cmake --build build ``` #### 4. Build Ollama binary: ``` go clean -cache ``` ``` go build -o ollama . ``` #### 5. Run ``` ./ollama serve ``` Since Ollama wasn't working, I was trying to use llama.cpp but then I had to compile that for my machine, but then I thought why not just compile ollama! Flags: - `--preset "ROCm 6"` - loads Ollama's predefined ROCm build configuration - `-DAMDGPU_TARGETS` / `-DGPU_TARGETS` - specifies your GPU architecture (both needed for ROCm compatibility) - `-DCMAKE_PREFIX_PATH` - tells CMake where to find your ROCm installation - `-Dhip_DIR` - points directly to the HIP cmake config file There may be a better way to do it, but for now this worked for me, and hopefully this could be helpful to others that were stuck!
GiteaMirror commented 2026-04-22 14:09:21 -05:00
Author
Owner
Copy Link

@Undermyth commented on GitHub (Mar 12, 2026):

@prabhdatnoor That's really works! I recompiled Ollama as stated in WSL Ubuntu 24.04, 9070 XT, and now Ollama can recognize and use the GPU correctly

<!-- gh-comment-id:4046556582 --> @Undermyth commented on GitHub (Mar 12, 2026): @prabhdatnoor That's really works! I recompiled Ollama as stated in WSL Ubuntu 24.04, 9070 XT, and now Ollama can recognize and use the GPU correctly
GiteaMirror commented 2026-04-22 14:09:22 -05:00
Author
Owner
Copy Link

@EvgeniySpinov commented on GitHub (Mar 18, 2026):

Have anyone tried running 0.18.x versions of ollama?

ROCm was updated to version 7, so hopefully 9070 XT might be there. I've looked into artefacts though and on ROCm side I see only kernels and tensors for up to gfx1151 which is Strix APU. gfx1201 is till missing. But, perhaps would work without kernels.

<!-- gh-comment-id:4083149591 --> @EvgeniySpinov commented on GitHub (Mar 18, 2026): Have anyone tried running 0.18.x versions of ollama? ROCm was updated to version 7, so hopefully 9070 XT might be there. I've looked into artefacts though and on ROCm side I see only kernels and tensors for up to gfx1151 which is Strix APU. gfx1201 is till missing. But, perhaps would work without kernels.
GiteaMirror commented 2026-04-22 14:09:23 -05:00
Author
Owner
Copy Link

@sintel-be commented on GitHub (Mar 19, 2026):

I'm trying to get the latest docker container (ollama/ollama:rocm) - which should be 0.18.2 with ROCm 7.2 - on a linux host to get to use my 9070XT GPU, but it constantly falls back to the CPU - I'm trying to work through the issue together with claude code, but no luck so far.

EDIT: after a couple of hours, got it to recognize the GPU, the problem was SElinux preventing the usage of the GPU by the container as mentioned here: https://github.com/ollama/ollama/blob/main/docs/gpu.mdx#container-permission

<!-- gh-comment-id:4090037330 --> @sintel-be commented on GitHub (Mar 19, 2026): I'm trying to get the latest docker container (ollama/ollama:rocm) - which should be 0.18.2 with ROCm 7.2 - on a linux host to get to use my 9070XT GPU, but it constantly falls back to the CPU - I'm trying to work through the issue together with claude code, but no luck so far. EDIT: after a couple of hours, got it to recognize the GPU, the problem was SElinux preventing the usage of the GPU by the container as mentioned here: https://github.com/ollama/ollama/blob/main/docs/gpu.mdx#container-permission
GiteaMirror commented 2026-04-22 14:09:24 -05:00
Author
Owner
Copy Link

@frbelotto commented on GitHub (Apr 9, 2026):

Did anyone get it working under a ollama docker image?

<!-- gh-comment-id:4210770538 --> @frbelotto commented on GitHub (Apr 9, 2026): Did anyone get it working under a ollama docker image?
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label feature request windows
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#32616
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?