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 #15482] ROCm on AMD Phoenix APUs (Radeon 780M iGPU) — HSA_OVERRIDE_GFX_VERSION required #35658

New Issue
Open
opened 2026-04-22 20:19:49 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-22 20:19:49 -05:00
Owner
Copy Link

Originally created by @rkocosmergon on GitHub (Apr 10, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15482

Sharing a finding that might help others running Ollama on AMD Ryzen 7000/8000 series with integrated graphics (Phoenix APUs, Radeon 780M).

Out of the box, Ollama doesn't detect the iGPU and falls back to CPU. The fix is one environment variable: export HSA_OVERRIDE_GFX_VERSION=11.0.0

This tells ROCm to treat the Phoenix APU as a supported gfx1100 target. Without it, the GPU sits idle — no error, no warning, just silent CPU fallback.

After setting this + installing rocm-libs:

  • llama3.2:3b: ~20 tok/s (was ~1 tok/s on CPU)
  • phi4-mini: 14-20 tok/s
  • 31.4 GiB unified memory available (shared system RAM)

Discovered this on a Hetzner AX42-U (Ryzen 7 PRO 8700GE) where the iGPU isn't even listed in the server specs. Full benchmark writeup: https://cosmergon.com/reports/llm-benchmark-hetzner.html

Might be worth adding a note about Phoenix APUs to the ROCm/AMD section of the Ollama docs?

Originally created by @rkocosmergon on GitHub (Apr 10, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15482 Sharing a finding that might help others running Ollama on AMD Ryzen 7000/8000 series with integrated graphics (Phoenix APUs, Radeon 780M). Out of the box, Ollama doesn't detect the iGPU and falls back to CPU. The fix is one environment variable: export HSA_OVERRIDE_GFX_VERSION=11.0.0 This tells ROCm to treat the Phoenix APU as a supported gfx1100 target. Without it, the GPU sits idle — no error, no warning, just silent CPU fallback. After setting this + installing rocm-libs: - llama3.2:3b: ~20 tok/s (was ~1 tok/s on CPU) - phi4-mini: 14-20 tok/s - 31.4 GiB unified memory available (shared system RAM) Discovered this on a Hetzner AX42-U (Ryzen 7 PRO 8700GE) where the iGPU isn't even listed in the server specs. Full benchmark writeup: https://cosmergon.com/reports/llm-benchmark-hetzner.html Might be worth adding a note about Phoenix APUs to the ROCm/AMD section of the Ollama docs?
GiteaMirror commented 2026-04-22 20:19:51 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Apr 10, 2026):

https://github.com/ollama/ollama/blob/main/docs/gpu.mdx#overrides-on-linux:~:text=Radeon%20RX%207600-,gfx1103,-Radeon%20780M

<!-- gh-comment-id:4224592163 --> @rick-github commented on GitHub (Apr 10, 2026): https://github.com/ollama/ollama/blob/main/docs/gpu.mdx#overrides-on-linux:~:text=Radeon%20RX%207600-,gfx1103,-Radeon%20780M
GiteaMirror commented 2026-04-22 20:19:51 -05:00
Author
Owner
Copy Link

@rkocosmergon commented on GitHub (Apr 10, 2026):

Thanks Rick — you're right, the GPU docs do list gfx1103 for the 780M. I missed that page during debugging.

The part that tripped me up was the full path from "Hetzner AX42-U with no GPU in specs" to "working ROCm inference": discovering the iGPU via lspci, installing rocm-libs, and finding the right HSA_OVERRIDE value. Leaving this here in case others search for a similar setup.

<!-- gh-comment-id:4224695529 --> @rkocosmergon commented on GitHub (Apr 10, 2026): Thanks Rick — you're right, the GPU docs do list gfx1103 for the 780M. I missed that page during debugging. The part that tripped me up was the full path from "Hetzner AX42-U with no GPU in specs" to "working ROCm inference": discovering the iGPU via lspci, installing rocm-libs, and finding the right HSA_OVERRIDE value. Leaving this here in case others search for a similar setup.
GiteaMirror commented 2026-04-22 20:19:52 -05:00
Author
Owner
Copy Link

@PureBlissAK commented on GitHub (Apr 18, 2026):

🤖 Automated Triage & Analysis Report

Issue: #15482
Analyzed: 2026-04-18T18:21:09.936776

Analysis

  • Type: unknown
  • Severity: medium
  • Components: unknown

Implementation Plan

  • Effort: medium
  • Steps:

This issue has been triaged and marked for implementation.

<!-- gh-comment-id:4274307832 --> @PureBlissAK commented on GitHub (Apr 18, 2026): <!-- ollama-issue-orchestrator:v1 issue:15482 --> ## 🤖 Automated Triage & Analysis Report **Issue**: #15482 **Analyzed**: 2026-04-18T18:21:09.936776 ### Analysis - **Type**: unknown - **Severity**: medium - **Components**: unknown ### Implementation Plan - **Effort**: medium - **Steps**: *This issue has been triaged and marked for implementation.*
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
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#35658
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?