github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-17 05:36:16 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #13788] All models run from GPU crash the server on load with various errors but run fine on CPU #55545

New Issue
Closed
opened 2026-04-29 09:23:09 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-29 09:23:09 -05:00
Owner
Copy Link

Originally created by @PatrickAmara on GitHub (Jan 20, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/13788

What is the issue?

Running ollama 0.14.2 on Arch Linux installed from the official repos (ollama and ollama-rocm), my GPU is AMD Radeon RX 6700 XT (I'm using HSA_OVERRIDE_GFX_VERSION=10.3.0 so ollama can detect it). Every model I try to run on my GPU, including those that previously worked, fail to run with different errors per model (but the same error for the same model). For example:

  • gemma3:270m-it-qat always throws Error: 500 Internal Server Error: do load request: Post "http://127.0.0.1:36955/load": EOF
  • deepseek-r1:1.5b and 7b always throw Error: 500 Internal Server Error: llama runner process has terminated: GGML_ASSERT(max_blocks_per_sm > 0) failed

All these models work fine loaded to my CPU instead. I picked up Ollama after a year or so of not using it, so I'm not sure which is the last working version I used, but I can verify deepseek-r1:7b ran on my GPU back then.

Logs were captured from server run with flags OLLAMA_DEBUG=2 HSA_OVERRIDE_GFX_VERSION=10.3.0

server_log_deepseek.txt
server_log_gemma.txt

Relevant log output

OS

Linux

GPU

AMD

CPU

AMD

Ollama version

0.14.2

Originally created by @PatrickAmara on GitHub (Jan 20, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/13788 ### What is the issue? Running ollama 0.14.2 on Arch Linux installed from the official repos (`ollama` and `ollama-rocm`), my GPU is AMD Radeon RX 6700 XT (I'm using `HSA_OVERRIDE_GFX_VERSION=10.3.0` so ollama can detect it). Every model I try to run on my GPU, including those that previously worked, fail to run with different errors per model (but the same error for the same model). For example: - gemma3:270m-it-qat always throws `Error: 500 Internal Server Error: do load request: Post "http://127.0.0.1:36955/load": EOF` - deepseek-r1:1.5b and 7b always throw `Error: 500 Internal Server Error: llama runner process has terminated: GGML_ASSERT(max_blocks_per_sm > 0) failed` All these models work fine loaded to my CPU instead. I picked up Ollama after a year or so of not using it, so I'm not sure which is the last working version I used, but I can verify deepseek-r1:7b ran on my GPU back then. Logs were captured from server run with flags `OLLAMA_DEBUG=2 HSA_OVERRIDE_GFX_VERSION=10.3.0` [server_log_deepseek.txt](https://github.com/user-attachments/files/24731719/server_log_deepseek.txt) [server_log_gemma.txt](https://github.com/user-attachments/files/24731720/server_log_gemma.txt) ### Relevant log output ```shell ``` ### OS Linux ### GPU AMD ### CPU AMD ### Ollama version 0.14.2
GiteaMirror added the bug label 2026-04-29 09:23:09 -05:00
GiteaMirror commented 2026-04-29 09:23:12 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 20, 2026):

What version of ROCm? If it's 7.*, try downgrading to 6.*. If that's not possible, try running Vulkan instead: install ollama-vulkan and set OLLAMA_VULKAN=1 OLLAMA_LLM_LIBRARY=vulkan.

<!-- gh-comment-id:3772039757 --> @rick-github commented on GitHub (Jan 20, 2026): What version of ROCm? If it's 7.\*, try downgrading to 6.\*. If that's not possible, try running Vulkan instead: install ollama-vulkan and set `OLLAMA_VULKAN=1` `OLLAMA_LLM_LIBRARY=vulkan`.
GiteaMirror commented 2026-04-29 09:23:13 -05:00
Author
Owner
Copy Link

@PatrickAmara commented on GitHub (Jan 20, 2026):

ollama-vulkan works indeed, thanks! I'm on ROCm 7.1.1 and can't downgrade to test with 6.*, so that will do for now. Is there any open issue about ROCm 7 not working so I can mark my issue as duplicate, or a relevant PR I can follow until it is compatible?

<!-- gh-comment-id:3773681352 --> @PatrickAmara commented on GitHub (Jan 20, 2026): ollama-vulkan works indeed, thanks! I'm on ROCm 7.1.1 and can't downgrade to test with 6.*, so that will do for now. Is there any open issue about ROCm 7 not working so I can mark my issue as duplicate, or a relevant PR I can follow until it is compatible?
GiteaMirror commented 2026-04-29 09:23:13 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 20, 2026):

There's an open llama.cpp issue. This is the first report I'm aware of that shows an impact in ollama.

<!-- gh-comment-id:3773770556 --> @rick-github commented on GitHub (Jan 20, 2026): There's an open [llama.cpp issue](https://github.com/ggml-org/llama.cpp/issues/17171). This is the first report I'm aware of that shows an impact in ollama.
GiteaMirror commented 2026-04-29 09:23:14 -05:00
Author
Owner
Copy Link

@nonetrix commented on GitHub (Jan 31, 2026):

Same here, flash attention on AMD is even a thing now? Is it just broken on older GPUs? I remember trying to get that to work on Stable Diffusion on my RX 6800 no luck

<!-- gh-comment-id:3827490752 --> @nonetrix commented on GitHub (Jan 31, 2026): Same here, flash attention on AMD is even a thing now? Is it just broken on older GPUs? I remember trying to get that to work on Stable Diffusion on my RX 6800 no luck
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#55545
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?