github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-15 20:54:54 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #8806] Empty supported_types list prevents GPU detection for older AMD GPUs (RX 5700 XT) in official Ollama distribution #67770

New Issue
Closed
opened 2026-05-04 11:38:24 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-04 11:38:24 -05:00
Owner
Copy Link

Originally created by @Molier on GitHub (Feb 4, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8806

What is the issue?

While trying to use Ollama with an AMD RX 5700 XT (gfx1010:xnack-), I've identified what appears to be an implementation issue in the official Ollama distribution's GPU detection system. This issue persists across multiple ROCm versions and setup approaches.

Core Issue:

The supported_types list is returning empty ([]), which prevents ANY GPU from being recognized, even though the ROCm stack successfully detects the GPU:

  1. hipinfo.exe properly detects the GPU:

    • Name: AMD RX 5700 XT
    • gcnArchName: gfx1010:xnack-

  2. Ollama log shows:

    • msg="amdgpu is not supported (supported types:[])"
    • Note the empty array for supported types

Investigation Steps Taken:

Technical Details:

  • Ollama version: 0.5.7
  • ROCm versions tested: 5.7, 6.1, 6.2.4
  • OS: Windows
  • Environment variables set: HSA_OVERRIDE_GFX_VERSION=10.1.0
  • Libraries path: C:\Users[user]\AppData\Local\Programs\Ollama\lib\ollama

Expected Behavior:

  • supported_types should contain a list of supported GPU architectures, even if my specific architecture isn't included
  • Other users have reported successful implementation with the same GPU model (see LinkedIn reference)

Actual Behavior:

  • supported_types is consistently returning an empty array ([])
  • This prevents the GPU detection system from functioning properly, regardless of ROCm version or library configuration

Request:
Could you please investigate:

  1. Why the supported_types list is empty in the official distribution
  2. What determines the population of this list
  3. Whether this might be related to how GPU architecture support is implemented

This information would be valuable for users trying to work with AMD GPUs, especially considering there are documented successful implementations with this exact GPU model.

exact log ( it shows this type of log with or without any of the rocms or any of the fixes from the git for unsupported rocm devices) this list just never populates supported types:[]:

PS C:\Users\user\AppData\Local\Programs\Ollama\lib> ollama serve
2025/02/04 00:56:19 routes.go:1187: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES:0 HSA_OVERRIDE_GFX_VERSION:10.1.0 HTTPS_PROXY: HTTP_PROXY: NO_PROXY: 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:\\Users\\user\\.ollama\\models OLLAMA_MULTIUSER_CACHE: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://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:0]"
time=2025-02-04T00:56:19.892+01:00 level=INFO source=images.go:432 msg="total blobs: 0"
time=2025-02-04T00:56:19.892+01:00 level=INFO source=images.go:439 msg="total unused blobs removed: 0"
time=2025-02-04T00:56:19.892+01:00 level=INFO source=routes.go:1238 msg="Listening on 127.0.0.1:11434 (version 0.5.7)"
time=2025-02-04T00:56:19.893+01:00 level=INFO source=routes.go:1267 msg="Dynamic LLM libraries" runners="[rocm_avx cpu cpu_avx cpu_avx2 cuda_v11_avx cuda_v12_avx]"
time=2025-02-04T00:56:19.893+01:00 level=INFO source=gpu.go:226 msg="looking for compatible GPUs"
time=2025-02-04T00:56:19.893+01:00 level=INFO source=gpu_windows.go:167 msg=packages count=1
time=2025-02-04T00:56:19.893+01:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16
time=2025-02-04T00:56:19.925+01:00 level=INFO source=amd_windows.go:72 msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=10.1.0
time=2025-02-04T00:56:20.290+01:00 level=WARN source=amd_windows.go:140 msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1010:xnack- gpu=0 library=C:\Users\user\AppData\Local\Programs\Ollama\lib\ollama
time=2025-02-04T00:56:20.292+01:00 level=INFO source=gpu.go:392 msg="no compatible GPUs were discovered"
time=2025-02-04T00:56:20.292+01:00 level=INFO source=types.go:131 msg="inference compute" id=0 library=cpu variant=avx2 compute="" driver=0.0 name="" total="31.9 GiB" available="19.6 GiB"

OS

Windows

GPU

AMD

CPU

AMD

Ollama version

0.5.7

Originally created by @Molier on GitHub (Feb 4, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8806 ### What is the issue? While trying to use Ollama with an AMD RX 5700 XT (gfx1010:xnack-), I've identified what appears to be an implementation issue in the official Ollama distribution's GPU detection system. This issue persists across multiple ROCm versions and setup approaches. # Core Issue: The supported_types list is returning empty ([]), which prevents ANY GPU from being recognized, even though the ROCm stack successfully detects the GPU: 1. hipinfo.exe properly detects the GPU: - Name: AMD RX 5700 XT - gcnArchName: gfx1010:xnack- 2. Ollama log shows: - msg="amdgpu is not supported (supported types:[])" - Note the empty array for supported types ## Investigation Steps Taken: - Tested with multiple ROCm versions (5.7, 6.1, 6.2.4) from AMD's official HIP SDK Reference: https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html - Tried various builds from ollama-for-amd repository Reference: https://github.com/likelovewant/ollama-for-amd - Followed successful implementation guides: - https://www.linkedin.com/pulse/ollama-working-amd-rx-5700-xt-windows-robert-buccigrossi-tze0e/ - https://medium.com/@afilipe362/using-ollama-on-older-amd-gpus-with-rocm-on-windows-a0b08de2cf02 - Used specific ROCm libraries for gfx1010: - https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/tag/v0.6.2.4 - https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/tag/v0.6.1.2 ## Technical Details: - Ollama version: 0.5.7 - ROCm versions tested: 5.7, 6.1, 6.2.4 - OS: Windows - Environment variables set: HSA_OVERRIDE_GFX_VERSION=10.1.0 - Libraries path: C:\Users\[user]\AppData\Local\Programs\Ollama\lib\ollama ### Expected Behavior: - supported_types should contain a list of supported GPU architectures, even if my specific architecture isn't included - Other users have reported successful implementation with the same GPU model (see LinkedIn reference) ### Actual Behavior: - supported_types is consistently returning an empty array ([]) - This prevents the GPU detection system from functioning properly, regardless of ROCm version or library configuration Request: Could you please investigate: 1. Why the supported_types list is empty in the official distribution 2. What determines the population of this list 3. Whether this might be related to how GPU architecture support is implemented This information would be valuable for users trying to work with AMD GPUs, especially considering there are documented successful implementations with this exact GPU model. exact log ( it shows this type of log with or without any of the rocms or any of the fixes from the git for unsupported rocm devices) this list just never populates supported types:[]: ``` PS C:\Users\user\AppData\Local\Programs\Ollama\lib> ollama serve 2025/02/04 00:56:19 routes.go:1187: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES:0 HSA_OVERRIDE_GFX_VERSION:10.1.0 HTTPS_PROXY: HTTP_PROXY: NO_PROXY: 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:\\Users\\user\\.ollama\\models OLLAMA_MULTIUSER_CACHE: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://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES:0]" time=2025-02-04T00:56:19.892+01:00 level=INFO source=images.go:432 msg="total blobs: 0" time=2025-02-04T00:56:19.892+01:00 level=INFO source=images.go:439 msg="total unused blobs removed: 0" time=2025-02-04T00:56:19.892+01:00 level=INFO source=routes.go:1238 msg="Listening on 127.0.0.1:11434 (version 0.5.7)" time=2025-02-04T00:56:19.893+01:00 level=INFO source=routes.go:1267 msg="Dynamic LLM libraries" runners="[rocm_avx cpu cpu_avx cpu_avx2 cuda_v11_avx cuda_v12_avx]" time=2025-02-04T00:56:19.893+01:00 level=INFO source=gpu.go:226 msg="looking for compatible GPUs" time=2025-02-04T00:56:19.893+01:00 level=INFO source=gpu_windows.go:167 msg=packages count=1 time=2025-02-04T00:56:19.893+01:00 level=INFO source=gpu_windows.go:214 msg="" package=0 cores=8 efficiency=0 threads=16 time=2025-02-04T00:56:19.925+01:00 level=INFO source=amd_windows.go:72 msg="skipping rocm gfx compatibility check" HSA_OVERRIDE_GFX_VERSION=10.1.0 time=2025-02-04T00:56:20.290+01:00 level=WARN source=amd_windows.go:140 msg="amdgpu is not supported (supported types:[])" gpu_type=gfx1010:xnack- gpu=0 library=C:\Users\user\AppData\Local\Programs\Ollama\lib\ollama time=2025-02-04T00:56:20.292+01:00 level=INFO source=gpu.go:392 msg="no compatible GPUs were discovered" time=2025-02-04T00:56:20.292+01:00 level=INFO source=types.go:131 msg="inference compute" id=0 library=cpu variant=avx2 compute="" driver=0.0 name="" total="31.9 GiB" available="19.6 GiB" ``` ### OS Windows ### GPU AMD ### CPU AMD ### Ollama version 0.5.7
GiteaMirror added the bug label 2026-05-04 11:38:24 -05:00
GiteaMirror commented 2026-05-04 11:38:28 -05:00
Author
Owner
Copy Link

@hamburger-114514 commented on GitHub (Feb 13, 2025):

I have the same issue, and my GPU is AMD RX 6750 XT

<!-- gh-comment-id:2656264441 --> @hamburger-114514 commented on GitHub (Feb 13, 2025): I have the same issue, and my GPU is AMD RX 6750 XT
GiteaMirror commented 2026-05-04 11:38:29 -05:00
Author
Owner
Copy Link

@Molier commented on GitHub (Feb 21, 2025):

I eventually fixed this by NOT setting any env varibles, i also just smashed the relevant files i needed from the rocm libraries by ilovekwant into all relative directoris i could find since most would not get detected.
rocm.gfx1010-xnack-gfx1011-xnack-gfx1012-xnack-.for.hip.sdk.6.2.4.7z was the files i downloaded but depends on gpu ofc
Concise Solution (Workaround for RX 5700 XT and similar):

GPU acceleration is now working with the official Ollama (v0.5.7) on Windows for my AMD RX 5700 XT. Here are the key steps:

  1. Get ROCm Libraries from likelovewant's Repo: Download rocm.gfx1010-xnack-gfx1011-xnack-gfx1012-xnack-.for.hip.sdk.6.2.4.7z from the v0.6.2.4 release at https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases.

  2. Extract and Copy "library" Folder Contents:

    • Extract the downloaded .7z archive.
    • From the extracted archive, copy the contents of the top-level "library" folder (and potentially the top-level "rocblas" folder if present in the archive, though "library" is key).
    • Paste these copied contents into these Ollama installation directories (you may need to create some if they don't exist):
      • C:\Users\%username%\AppData\Local\Programs\Ollama\lib\ollama\
      • C:\Users\%username%\AppData\Local\Programs\Ollama\lib\ollama\library\
      • C:\Users\%username%\AppData\Local\Programs\Ollama\lib\ollama\rocblas\
      • C:\Users\%username%\AppData\Local\Programs\Ollama\rocm\

  3. Important: Remove Environment Variables: If you previously set HSA_OVERRIDE_GFX_VERSION or HCC_AMDGPU_TARGET environment variables, remove them. For me, these were causing issues.

  4. ROCm SDK Installed: I had the official ROCm SDK 6.2 installed at C:\Program Files\AMD\ROCm\6.2\. This might be necessary.

Quick Advice & Repo Link:

  • Primary Resource: Use https://github.com/likelovewant/ollama-for-amd as your main guide for unsupported AMD GPUs.
  • Experiment: Trial and error might be needed. Library placement and environment variables can be system-dependent.
  • Check Logs: Use ollama serve and check the console output for errors.

Disclaimer: This is an unofficial workaround and may not be stable or work for everyone.

I generally recommend going to this repo, https://github.com/likelovewant/ollama-for-amd and following steps there as well, but i ended up getting it workign on official release.

question

id still like to know exactly where that ollama looks for its libs, since if that would have been in the logs, a lot of pain would have been spared. so some more verbose logging would be nice.

<!-- gh-comment-id:2675076628 --> @Molier commented on GitHub (Feb 21, 2025): I eventually fixed this by NOT setting any env varibles, i also just smashed the relevant files i needed from the rocm libraries by ilovekwant into all relative directoris i could find since most would not get detected. ` rocm.gfx1010-xnack-gfx1011-xnack-gfx1012-xnack-.for.hip.sdk.6.2.4.7z ` was the files i downloaded but depends on gpu ofc **Concise Solution (Workaround for RX 5700 XT and similar):** GPU acceleration is now working with the official Ollama (v0.5.7) on Windows for my AMD RX 5700 XT. Here are the key steps: 1. **Get ROCm Libraries from likelovewant's Repo:** Download `rocm.gfx1010-xnack-gfx1011-xnack-gfx1012-xnack-.for.hip.sdk.6.2.4.7z` from the [v0.6.2.4 release](https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/tag/v0.6.2.4) at [https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases](https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases). 2. **Extract and Copy "library" Folder Contents:** * Extract the downloaded `.7z` archive. * **From the extracted archive, copy the *contents* of the top-level "library" folder** (and potentially the top-level "rocblas" folder if present in the archive, though "library" is key). * Paste these copied contents into these **Ollama installation directories** (you may need to create some if they don't exist): * `C:\Users\%username%\AppData\Local\Programs\Ollama\lib\ollama\` * `C:\Users\%username%\AppData\Local\Programs\Ollama\lib\ollama\library\` * `C:\Users\%username%\AppData\Local\Programs\Ollama\lib\ollama\rocblas\` * `C:\Users\%username%\AppData\Local\Programs\Ollama\rocm\` 3. **Important: Remove Environment Variables:** If you previously set `HSA_OVERRIDE_GFX_VERSION` or `HCC_AMDGPU_TARGET` environment variables, **remove them.** For me, these were causing issues. 4. **ROCm SDK Installed:** I had the official **ROCm SDK 6.2** installed at `C:\Program Files\AMD\ROCm\6.2\`. This might be necessary. **Quick Advice & Repo Link:** * **Primary Resource:** Use [https://github.com/likelovewant/ollama-for-amd](https://github.com/likelovewant/ollama-for-amd) as your main guide for unsupported AMD GPUs. * **Experiment:** Trial and error might be needed. Library placement and environment variables can be system-dependent. * **Check Logs:** Use `ollama serve` and check the console output for errors. **Disclaimer:** This is an unofficial workaround and may not be stable or work for everyone. I generally recommend going to this repo, https://github.com/likelovewant/ollama-for-amd and following steps there as well, but i ended up getting it workign on official release. # question id still like to know exactly where that ollama looks for its libs, since if that would have been in the logs, a lot of pain would have been spared. so some more verbose logging would be nice.
GiteaMirror commented 2026-05-04 11:38:30 -05:00
Author
Owner
Copy Link

@aleksei-melinic commented on GitHub (Oct 13, 2025):

@Molier Thank you, worked for me ❤️

<!-- gh-comment-id:3397523906 --> @aleksei-melinic commented on GitHub (Oct 13, 2025): @Molier Thank you, worked for me ❤️
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 bug
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#67770
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?