github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[PR #13477] Rocm igpu uma #24765

New Issue
Open
opened 2026-04-19 17:47:51 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-19 17:47:51 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/13477
Author: @marx161-cmd
Created: 12/15/2025
Status: 🔄 Open

Base: main ← Head: rocm-igpu-uma

📝 Commits (7)

  • f8e4c81 Add experimental ROCm iGPU support
  • 598c08d Add ROCm UMA run log
  • 7e2aac6 Add OLLAMA_NUM_GPU default
  • 9582fea Add AMD iGPU fork banner
  • 8606cb5 Document AMD iGPU fork differences
  • 368928b rocm: fix HIP clang conflicts and disable MMF
  • ba051e0 docs: note ROCm 7.1 HIP build fixes

📊 Changes

12 files changed (+914 additions, -28 deletions)

View changed files

📝 README.md (+17 -0)
📝 api/types.go (+1 -1)
➕ cmake/CMakeDetermineHIPCompiler.cmake (+329 -0)
➕ docs/logs/rocm-760m-run-response.json (+1 -0)
➕ docs/logs/rocm-760m-run.log (+1 -0)
➕ docs/rocm-apu.md (+92 -0)
📝 envconfig/config.go (+19 -0)
📝 llm/server.go (+30 -0)
📝 ml/backend/ggml/ggml/src/ggml-cuda/mmf.cu (+11 -0)
📝 ml/backend/ggml/ggml/src/ggml-cuda/vendors/hip.h (+330 -0)
📝 ml/backend/ggml/ggml/src/mem_hip.cpp (+79 -27)
📝 ml/device.go (+4 -0)

📄 Description

Summary

  • Enable HIP builds with staged ROCm runtimes so UMA-only AMD APUs (e.g. Radeon 760M) can run Ollama.

  • Adjust runner/device logic to respect user-provided OLLAMA_LIBRARY_PATH entries and skip incompatibility filtering when HSA_OVERRIDE_GFX_VERSION is already set.

  • Document the full setup (staging RPMs, build flags, runtime env vars) and include a sample discovery log.

Testing

  • Fedora 43, AMD Radeon 760M iGPU (UMA 16 GiB)

  • ./build/ollama serve with:

HSA_OVERRIDE_GFX_VERSION=11.0.0
GGML_HIP_FORCE_GTT=1
OLLAMA_LIBRARY_PATH=$PWD/build/lib/ollama/rocm:$PWD/build/lib/ollama
LD_LIBRARY_PATH=$OLLAMA_LIBRARY_PATH:$PWD/build/rocm-stage/lib64

  • curl http://127.0.0.1:11434/api/generate -d '{"model":"phi-3.5-mini","prompt":"Say ROCm test","stream":false}'

  • Logs: docs/logs/rocm-760m-run.log

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/ollama/ollama/pull/13477 **Author:** [@marx161-cmd](https://github.com/marx161-cmd) **Created:** 12/15/2025 **Status:** 🔄 Open **Base:** `main` ← **Head:** `rocm-igpu-uma` --- ### 📝 Commits (7) - [`f8e4c81`](https://github.com/ollama/ollama/commit/f8e4c812b9f89e2e4fb8cc6f7b0657798c5dc8a1) Add experimental ROCm iGPU support - [`598c08d`](https://github.com/ollama/ollama/commit/598c08d22b7b234ceb8222a7c216e39e4d5dd5d4) Add ROCm UMA run log - [`7e2aac6`](https://github.com/ollama/ollama/commit/7e2aac6255cb80e0a347f7022dc025e315d23f77) Add OLLAMA_NUM_GPU default - [`9582fea`](https://github.com/ollama/ollama/commit/9582feac7c1bf4d01d8a3279ac25c2aafe79a2d2) Add AMD iGPU fork banner - [`8606cb5`](https://github.com/ollama/ollama/commit/8606cb5acf01796a741bf008a279aa45dc1985e9) Document AMD iGPU fork differences - [`368928b`](https://github.com/ollama/ollama/commit/368928bdad0c3d7f9e26f534a74d6a82b5933ab9) rocm: fix HIP clang conflicts and disable MMF - [`ba051e0`](https://github.com/ollama/ollama/commit/ba051e042a182dcda552d429fc1ae12cdd0807da) docs: note ROCm 7.1 HIP build fixes ### 📊 Changes **12 files changed** (+914 additions, -28 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+17 -0) 📝 `api/types.go` (+1 -1) ➕ `cmake/CMakeDetermineHIPCompiler.cmake` (+329 -0) ➕ `docs/logs/rocm-760m-run-response.json` (+1 -0) ➕ `docs/logs/rocm-760m-run.log` (+1 -0) ➕ `docs/rocm-apu.md` (+92 -0) 📝 `envconfig/config.go` (+19 -0) 📝 `llm/server.go` (+30 -0) 📝 `ml/backend/ggml/ggml/src/ggml-cuda/mmf.cu` (+11 -0) 📝 `ml/backend/ggml/ggml/src/ggml-cuda/vendors/hip.h` (+330 -0) 📝 `ml/backend/ggml/ggml/src/mem_hip.cpp` (+79 -27) 📝 `ml/device.go` (+4 -0) </details> ### 📄 Description ## Summary - Enable HIP builds with staged ROCm runtimes so UMA-only AMD APUs (e.g. Radeon 760M) can run Ollama. - Adjust runner/device logic to respect user-provided `OLLAMA_LIBRARY_PATH` entries and skip incompatibility filtering when `HSA_OVERRIDE_GFX_VERSION` is already set. - Document the full setup (staging RPMs, build flags, runtime env vars) and include a sample discovery log. ## Testing - Fedora 43, AMD Radeon 760M iGPU (UMA 16 GiB) - `./build/ollama serve` with: HSA_OVERRIDE_GFX_VERSION=11.0.0 GGML_HIP_FORCE_GTT=1 OLLAMA_LIBRARY_PATH=$PWD/build/lib/ollama/rocm:$PWD/build/lib/ollama LD_LIBRARY_PATH=$OLLAMA_LIBRARY_PATH:$PWD/build/rocm-stage/lib64 - `curl http://127.0.0.1:11434/api/generate -d '{"model":"phi-3.5-mini","prompt":"Say ROCm test","stream":false}'` - Logs: docs/logs/rocm-760m-run.log --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
GiteaMirror added the pull-request label 2026-04-19 17:47:51 -05:00
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 pull-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#24765
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?