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 #12359] Qwen3 crashes on gfx1150 iGPU #54724

New Issue
Open
opened 2026-04-29 07:05:52 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-29 07:05:52 -05:00
Owner
Copy Link

Originally created by @isapir on GitHub (Sep 20, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12359

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

ollama run qwen3 crashes on a simple question like "Why is the sky blue?". Gemma3 answers that question on the same machine.

ollama-qwen3-crash.log

Relevant log output

Attached file

OS

Linux

GPU

AMD

CPU

AMD

Ollama version

0.11.11

Originally created by @isapir on GitHub (Sep 20, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12359 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? `ollama run qwen3` crashes on a simple question like "Why is the sky blue?". Gemma3 answers that question on the same machine. [ollama-qwen3-crash.log](https://github.com/user-attachments/files/22447695/ollama-qwen3-crash.log) ### Relevant log output ```shell Attached file ``` ### OS Linux ### GPU AMD ### CPU AMD ### Ollama version 0.11.11
GiteaMirror added the amdbug labels 2026-04-29 07:05:54 -05:00
GiteaMirror commented 2026-04-29 07:05:56 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Sep 20, 2025):

@dhiltgen possible to take a quick look? Seems like a ROCm error on a gfx1150 device

<!-- gh-comment-id:3315305402 --> @jmorganca commented on GitHub (Sep 20, 2025): @dhiltgen possible to take a quick look? Seems like a ROCm error on a gfx1150 device
GiteaMirror commented 2026-04-29 07:05:57 -05:00
Author
Owner
Copy Link

@isapir commented on GitHub (Sep 20, 2025):

@jmorganca You think that this is a gfx1150 issue? How come gemma3 works without crashing? Or at least without crashing immediately - it did crash after some usage

A few days ago I reported https://github.com/ollama/ollama/issues/12320 which seems to be related given your comment

<!-- gh-comment-id:3315326682 --> @isapir commented on GitHub (Sep 20, 2025): @jmorganca You think that this is a `gfx1150` issue? How come `gemma3` works without crashing? Or at least without crashing immediately - it did crash after some usage A few days ago I reported https://github.com/ollama/ollama/issues/12320 which seems to be related given your comment
GiteaMirror commented 2026-04-29 07:05:57 -05:00
Author
Owner
Copy Link

@kekxv commented on GitHub (Sep 26, 2025):

Character error problem
Image

<!-- gh-comment-id:3336683933 --> @kekxv commented on GitHub (Sep 26, 2025): Character error problem <img width="601" height="370" alt="Image" src="https://github.com/user-attachments/assets/e1b6ee2b-7a03-4a16-9983-a3b78689d6ce" />
GiteaMirror commented 2026-04-29 07:05:58 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Sep 26, 2025):

ROCm does not support gfx1150, only gfx1151. That said, on Linux, I've yet to see a setup where you can dedicate more system memory to this iGPU. Our current logic requires dedicated VRAM. On windows with the AMD Adrenalin Edition software, you can adjust the Performance->Tuning to adjust how much RAM is dedicated to VRAM. But there's a catch. ROCm on windows doesn't support HSA_OVERRIDE_GFX_VERSION so even then, you can't get gfx1150 to mimic gfx1151 so ROCm will work on it.

From the logs, we were going to use CPU inference, but for some reason we're still loading ROCm and it's trying to initialize the card even though we set 0 layers to load, and that's causing a crash due to the unsupported gfx type. As a workaround to get inference working on the CPU on this system, set ROCR_VISIBLE_DEVICES=-1 to disable ROCm from attempting to work with the GPU until we can get that fixed.

<!-- gh-comment-id:3339266433 --> @dhiltgen commented on GitHub (Sep 26, 2025): ROCm does not support gfx1150, only gfx1151. That said, on Linux, I've yet to see a setup where you can dedicate more system memory to this iGPU. Our current logic requires dedicated VRAM. On windows with the AMD Adrenalin Edition software, you can adjust the Performance->Tuning to adjust how much RAM is dedicated to VRAM. But there's a catch. ROCm on windows doesn't support HSA_OVERRIDE_GFX_VERSION so even then, you can't get gfx1150 to mimic gfx1151 so ROCm will work on it. From the logs, we were going to use CPU inference, but for some reason we're still loading ROCm and it's trying to initialize the card even though we set 0 layers to load, and that's causing a crash due to the unsupported gfx type. As a workaround to get inference working on the CPU on this system, set `ROCR_VISIBLE_DEVICES=-1` to disable ROCm from attempting to work with the GPU until we can get that fixed.
GiteaMirror commented 2026-04-29 07:05:58 -05:00
Author
Owner
Copy Link

@rkkoszewski commented on GitHub (Oct 14, 2025):

@dhiltgen ROCm 6.4.4 does support gfx1150 officially now. The ROCm version would probably need to be updated on the Ollama side.

<!-- gh-comment-id:3401609094 --> @rkkoszewski commented on GitHub (Oct 14, 2025): @dhiltgen ROCm 6.4.4 does support gfx1150 officially now. The ROCm version would probably need to be updated on the Ollama side.
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 amd 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#54724
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?