github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10152] RX580 #32420

New Issue
Closed
opened 2026-04-22 13:40:06 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-22 13:40:06 -05:00
Owner
Copy Link

Originally created by @dpblnt on GitHub (Apr 6, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10152

I just spend some hours to build ollama for RX580, which is disabled on purpose.
ollama works perfectly with rocBLAS-5.7.1-r2 on gfx803
Why do all that work to disable it if it works?

Originally created by @dpblnt on GitHub (Apr 6, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10152 I just spend some hours to build ollama for RX580, which is disabled on purpose. ollama works perfectly with rocBLAS-5.7.1-r2 on gfx803 Why do all that work to disable it if it works?
GiteaMirror added the feature request label 2026-04-22 13:40:06 -05:00
GiteaMirror commented 2026-04-22 13:40:59 -05:00
Author
Owner
Copy Link

@Abdulhanan535 commented on GitHub (Apr 7, 2025):

I was about to build it, glad I checked.

<!-- gh-comment-id:2782175429 --> @Abdulhanan535 commented on GitHub (Apr 7, 2025): I was about to build it, glad I checked.
GiteaMirror commented 2026-04-22 13:40:59 -05:00
Author
Owner
Copy Link

@robertrosenbusch commented on GitHub (Apr 13, 2025):

@dpblnt @Abdulhanan535 : Just a hint.... RX570/580/590 are all the same on Polaris/gfx803 family.

on discover/gpu.go on line #73 you should change on Ollama v0.6.5.

var RocmComputeMajorMin = "9" to var RocmComputeMajorMin = "8" (someone decided to not support gfx803 on Ollama anymore and wanna support abroad gfx9xx^.^)

  • This short change on Ollama SourceCode could be save lifetime on gfx803 :P

The next thing:

  • gfx803 isnt a part to compile on ollama/CMakePresets.json on line #59
  • gfx803 isnt a part to compile on ollama//CMakeLists.txt on line #100

After this short changes and adds.. you are able to (re-)compile your RX5X0 aka gfx803 from git and use without any probs on good well known working ROCm :P

Just a hint, it works fine on ROCm 6.3 and recompile rocBLAS, but beware, there are some issues on Kernel 6.12 and 6.13 to use /dev/kfd and /dev/dri . But i guess i repeat myself again and again ^.^

Cheers, Robert Rosenbusch

<!-- gh-comment-id:2800113943 --> @robertrosenbusch commented on GitHub (Apr 13, 2025): @dpblnt @Abdulhanan535 : Just a hint.... RX570/580/590 are all the same on Polaris/gfx803 family. on `discover/gpu.go` on[ line #73 ](https://github.com/ollama/ollama/blob/main/discover/gpu.go) you should change on Ollama v0.6.5. `var RocmComputeMajorMin = "9"` to `var RocmComputeMajorMin = "8"` (**someone decided** to not support [gfx803 on Ollama anymore ](https://github.com/likelovewant/ollama-for-amd/issues/51 )and wanna support abroad gfx9xx^.^) - This short change on Ollama SourceCode could be save lifetime on gfx803 :P The next thing: - `gfx803` isnt a part to compile on `ollama/CMakePresets.json` on [line #59](https://github.com/ollama/ollama/blob/main/CMakePresets.json) - `gfx803` isnt a part to compile on `ollama//CMakeLists.txt` on line [#100](https://github.com/ollama/ollama/blob/main/CMakeLists.txt) After this short changes and adds.. you are able to (re-)compile your RX5X0 aka gfx803 from git _and_ use without any probs on good well known working ROCm :P Just a hint, it works fine on ROCm 6.3 and recompile rocBLAS, but beware, there are some issues on Kernel 6.12 and 6.13 to use `/dev/kfd` and `/dev/dri` . But i guess i repeat myself again and again ^.^ --- Cheers, Robert Rosenbusch
GiteaMirror commented 2026-04-22 13:41:00 -05:00
Author
Owner
Copy Link

@dpblnt commented on GitHub (Apr 14, 2025):

I was not able to make it work on rocm6.1, so I changed back to 5.7
As you mention RocmComputeMajorMin has to be set from 9 to lower.
I also had to disable some rocm detection in discover/amd_linux.go
then I build with

export CGO_LDFLAGS="-L$(pwd)/build3/lib/ollama -L$(pwd)/build3/ml/backend/ggml/ggml/src/ggml-hip -lggml-hip -Wl,-rpath,$(pwd)/build3/lib/ollama"
export CGO_CFLAGS="-I$(pwd)/build3 -I$(pwd)/ml/backend/ggml/ggml/include"
export CXX=/usr/bin/hipcc
export CC=/usr/lib/llvm/17/bin/clang
export CGO_ENABLED=1

rm -rf build3
mkdir build3
cd build3

cmake ..  -G Ninja -DCMAKE_C_COMPILER=clang DCMAKE_CXX_COMPILER=clang++  -DAMDGPU_TARGETS=gfx803
cmake --build . --config Release
cd ..
go clean -cache
go generate ./..
go build -o ollama

and run with

export ROC_ENABLE_PRE_VEGA=1
export HSA_OVERRIDE_GFX_VERSION="8.0.3"
export LD_LIBRARY_PATH=$PA/build3/lib/ollama
<!-- gh-comment-id:2800565525 --> @dpblnt commented on GitHub (Apr 14, 2025): I was not able to make it work on rocm6.1, so I changed back to 5.7 As you mention RocmComputeMajorMin has to be set from 9 to lower. I also had to disable some rocm detection in discover/amd_linux.go then I build with ``` export CGO_LDFLAGS="-L$(pwd)/build3/lib/ollama -L$(pwd)/build3/ml/backend/ggml/ggml/src/ggml-hip -lggml-hip -Wl,-rpath,$(pwd)/build3/lib/ollama" export CGO_CFLAGS="-I$(pwd)/build3 -I$(pwd)/ml/backend/ggml/ggml/include" export CXX=/usr/bin/hipcc export CC=/usr/lib/llvm/17/bin/clang export CGO_ENABLED=1 rm -rf build3 mkdir build3 cd build3 cmake .. -G Ninja -DCMAKE_C_COMPILER=clang DCMAKE_CXX_COMPILER=clang++ -DAMDGPU_TARGETS=gfx803 cmake --build . --config Release cd .. go clean -cache go generate ./.. go build -o ollama ``` and run with ``` export ROC_ENABLE_PRE_VEGA=1 export HSA_OVERRIDE_GFX_VERSION="8.0.3" export LD_LIBRARY_PATH=$PA/build3/lib/ollama ```
GiteaMirror commented 2026-04-22 13:41:00 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jul 5, 2025):

Old AMD GPU support is tracked via #2453

<!-- gh-comment-id:3040275358 --> @dhiltgen commented on GitHub (Jul 5, 2025): Old AMD GPU support is tracked via #2453
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
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#32420
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?