github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-15 12:48:05 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #12111] After ROCm Upgrade, HSA_OVERRIDE_GFX_VERSION to 10.3.0 Causes SIGSEGV After Receiving Prompt; No Override Works, But Falls Back to CPU #70107

New Issue
Open
opened 2026-05-04 20:22:12 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-05-04 20:22:12 -05:00
Owner
Copy Link

Originally created by @reteov on GitHub (Aug 29, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12111

What is the issue?

I'm using a gfx1031 GPU (Radeon RX 6750 XT), I just set the HSA_OVERRIDE_GFX_VERSION to 10.3.0 to allow ollama to work. It's been working fine for a while now, and I've not had any problems—at least, nothing that doesn't require more hardware to solve, and therefore has nothing to do with ollama.

However, a couple days ago, I pulled the trigger on a system upgrade. Among the upgrades were ones for ROCm (to 6.4.3). Once these upgrades were complete, Ollama would fail with a SIGSEGV whenever I submit a prompt, regardless if it's the command-line tool or the API. Removing the override would fix the problem, but return the LLM to CPU-only operation. I recently discovered that downgrading ROCm back to 6.4.1 also worked. However, obviously, while I can keep things as they are, it's not a long-term solution, and I'm not sure if this is something to approach ROCm ro you guys about, so I figured I'd place the bug report here first, in case you need to coordinate or something.

I've done some research, including with ChatGPT (itasked for some tools' output). I've included all the logs and reports I've made, and for anyone else who's having the same issue, downgrading ROCm back to 6.4.1 (at least, that I'm aware of) should allow GPU functionality to work again.

Please let me know if there's anything else you need.

Relevant log output

hardinfo2_report.html
installed.log ($ pacman -Q)
ollama-default.log (Log when not using HSA_OVERRIDE_GFX_VERSION)
ollama-gfx1030.log (Log when using HSA_OVERRIDE_GFX_VERSION)
ollama-symlink-addition.log (Log when using a symlink from gfx1030 to gfx1031)
rocminfo.log

OS

Arch Linux (it's a rolling release, so no specific version)

GPU

Radeon RX 6750 XT

CPU

AMD Ryzen 5 1600

Ollama version

ollama version is 0.11.7

Originally created by @reteov on GitHub (Aug 29, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12111 ### What is the issue? I'm using a gfx1031 GPU (Radeon RX 6750 XT), I just set the HSA_OVERRIDE_GFX_VERSION to 10.3.0 to allow ollama to work. It's been working fine for a while now, and I've not had any problems—at least, nothing that doesn't require more hardware to solve, and therefore has nothing to do with ollama. However, a couple days ago, I pulled the trigger on a system upgrade. Among the upgrades were ones for ROCm (to 6.4.3). Once these upgrades were complete, Ollama would fail with a SIGSEGV whenever I submit a prompt, regardless if it's the command-line tool or the API. Removing the override would fix the problem, but return the LLM to CPU-only operation. I recently discovered that downgrading ROCm back to 6.4.1 also worked. However, obviously, while I can keep things as they are, it's not a long-term solution, and I'm not sure if this is something to approach ROCm ro you guys about, so I figured I'd place the bug report here first, in case you need to coordinate or something. I've done some research, including with ChatGPT (itasked for some tools' output). I've included all the logs and reports I've made, and for anyone else who's having the same issue, downgrading ROCm back to 6.4.1 (at least, that I'm aware of) should allow GPU functionality to work again. Please let me know if there's anything else you need. ### Relevant log output [hardinfo2_report.html](https://github.com/user-attachments/files/22035275/hardinfo2_report.html) [installed.log](https://github.com/user-attachments/files/22035278/installed.log) (`$ pacman -Q`) [ollama-default.log](https://github.com/user-attachments/files/22035272/ollama-default.log) (Log when not using HSA_OVERRIDE_GFX_VERSION) [ollama-gfx1030.log](https://github.com/user-attachments/files/22035276/ollama-gfx1030.log) (Log when using HSA_OVERRIDE_GFX_VERSION) [ollama-symlink-addition.log](https://github.com/user-attachments/files/22035279/ollama-symlink-addition.log) (Log when using a symlink from gfx1030 to gfx1031) [rocminfo.log](https://github.com/user-attachments/files/22035277/rocminfo.log) ### OS Arch Linux (it's a rolling release, so no specific version) ### GPU Radeon RX 6750 XT ### CPU AMD Ryzen 5 1600 ### Ollama version ollama version is 0.11.7
GiteaMirror added the bug label 2026-05-04 20:22:14 -05:00
GiteaMirror commented 2026-05-04 20:22:23 -05:00
Author
Owner
Copy Link

@vcottineau commented on GitHub (Aug 30, 2025):

Hello, I am facing a similar issue since the latest Manjaro update.

Error

Error: model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details

août 30 20:17:58 manjaro ollama[1948]: time=2025-08-30T20:17:58.476+02:00 level=INFO source=server.go:637 msg="llama runner started in 11.04 seconds"
août 30 20:17:58 manjaro ollama[1948]: [GIN] 2025/08/30 - 20:17:58 | 200 | 11.071629851s |       127.0.0.1 | POST     "/api/generate"
août 30 20:18:00 manjaro ollama[1948]: SIGSEGV: segmentation violation

Logs

rocminfo.log
server.log

Version

CPU = AMD Ryzen 9 8945HS
GPU = AMD Radeon 780M
ROCm = 6.4.3
Ollama = 0.11.4

<!-- gh-comment-id:3239456603 --> @vcottineau commented on GitHub (Aug 30, 2025): Hello, I am facing a similar issue since the latest Manjaro update. ### Error ``` Error: model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details août 30 20:17:58 manjaro ollama[1948]: time=2025-08-30T20:17:58.476+02:00 level=INFO source=server.go:637 msg="llama runner started in 11.04 seconds" août 30 20:17:58 manjaro ollama[1948]: [GIN] 2025/08/30 - 20:17:58 | 200 | 11.071629851s | 127.0.0.1 | POST "/api/generate" août 30 20:18:00 manjaro ollama[1948]: SIGSEGV: segmentation violation ``` ### Logs [rocminfo.log](https://github.com/user-attachments/files/22060103/rocminfo.log) [server.log](https://github.com/user-attachments/files/22060114/server.log) ### Version CPU = AMD Ryzen 9 8945HS GPU = AMD Radeon 780M ROCm = 6.4.3 Ollama = 0.11.4
GiteaMirror commented 2026-05-04 20:22:25 -05:00
Author
Owner
Copy Link

@vcottineau commented on GitHub (Aug 31, 2025):

Today's ROCm update to 6.4.3-3 fixed the issue for me.

<!-- gh-comment-id:3240200062 --> @vcottineau commented on GitHub (Aug 31, 2025): Today's ROCm update to 6.4.3-3 fixed the issue for me.
GiteaMirror commented 2026-05-04 20:22:25 -05:00
Author
Owner
Copy Link

@apeacock commented on GitHub (Aug 31, 2025):

Where did you get the 6.4.3-3 update from, can't seem to find it.

<!-- gh-comment-id:3240439599 --> @apeacock commented on GitHub (Aug 31, 2025): Where did you get the 6.4.3-3 update from, can't seem to find it.
GiteaMirror commented 2026-05-04 20:22:26 -05:00
Author
Owner
Copy Link

@reteov commented on GitHub (Sep 1, 2025):

If you're using Arch, you'll need to make sure you're using sudo pacman -Syu to update your package list. If it is still not showing up, then you might need to refresh your mirror list.

Something to keep in mind is that it's RocBLAS that seemed to be causing the issue, and that is what's been updated to 6.4.3-3 (although rocsparse also is now at 6.4.3-3, as well)

I'm able to confirm that the issue is resolved.

<!-- gh-comment-id:3243059830 --> @reteov commented on GitHub (Sep 1, 2025): If you're using Arch, you'll need to make sure you're using `sudo pacman -Syu` to update your package list. If it is still not showing up, then you might need to refresh your mirror list. Something to keep in mind is that it's RocBLAS that seemed to be causing the issue, and that is what's been updated to 6.4.3-3 (although rocsparse also is now at 6.4.3-3, as well) I'm able to confirm that the issue is resolved.
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#70107
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?