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 #11299] Performance regression: v0.9.2 and higher #53966

New Issue
Open
opened 2026-04-29 05:01:17 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-29 05:01:17 -05:00
Owner
Copy Link

Originally created by @Stogas on GitHub (Jul 4, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11299

What is the issue?

Ollama v0.9.2, at least on Windows, takes a significantly more time before each inference starts on the GPU.

The following system usage graphs are from identical payload sequences and contents - 5 chat completions and 2 generations.

The issue persists across multiple reinstalls of each version. v0.9.1 and v0.9.0 are not affected.

CPU/GPU usage v0.9.1 - OK

Image Image

CPU/GPU usage v0.9.2 - REGRESSION

Image Image

CPU/GPU usage v0.9.5 - REGRESSION (same)

Image Image

Relevant log output

Normal:

Debug logs enabled:

Tracing enabled:

OS

Windows

GPU

AMD

CPU

AMD

Ollama version

0.9.2 and up

Originally created by @Stogas on GitHub (Jul 4, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11299 ### What is the issue? Ollama v0.9.2, at least on Windows, takes a significantly more time before each inference starts on the GPU. The following system usage graphs are from identical payload sequences and contents - 5 chat completions and 2 generations. The issue persists across multiple reinstalls of each version. v0.9.1 and v0.9.0 are not affected. <details><summary>CPU/GPU usage v0.9.1 - OK</summary> <p> <img width="540" height="468" alt="Image" src="https://github.com/user-attachments/assets/0646bc7a-e397-4033-bb69-8ad0a103b394" /> <img width="540" height="468" alt="Image" src="https://github.com/user-attachments/assets/00fd558f-7d8d-433b-95c0-cdbfe12a7536" /> </p> </details> <details><summary>CPU/GPU usage v0.9.2 - REGRESSION</summary> <p> <img width="540" height="468" alt="Image" src="https://github.com/user-attachments/assets/a29659b1-c53a-4548-bc20-766da4b46b9c" /> <img width="540" height="468" alt="Image" src="https://github.com/user-attachments/assets/e0004bbc-2118-4399-a799-4dfdf55b3920" /> </p> </details> <details><summary>CPU/GPU usage v0.9.5 - REGRESSION (same)</summary> <p> <img width="540" height="468" alt="Image" src="https://github.com/user-attachments/assets/bc6f0efc-7899-47fd-919d-c425966e2396" /> <img width="540" height="468" alt="Image" src="https://github.com/user-attachments/assets/6277d993-0155-469d-a47e-c5fd6e23e739" /> </p> </details> ### Relevant log output Normal: - [server-0.9.1.log](https://github.com/user-attachments/files/21064293/server-0.9.1.log) - [server-0.9.2.log](https://github.com/user-attachments/files/21064295/server-0.9.2.log) - [server-0.9.5.log](https://github.com/user-attachments/files/21064297/server-0.9.5.log) Debug logs enabled: - [server-0.9.1-debug.log](https://github.com/user-attachments/files/21064418/server-0.9.1-debug.log) - [server-0.9.2-debug.log](https://github.com/user-attachments/files/21064390/server-0.9.2-debug.log) - [server-0.9.5-debug.log](https://github.com/user-attachments/files/21064391/server-0.9.5-debug.log) Tracing enabled: - [server-0.9.1-trace.log](https://github.com/user-attachments/files/21064567/server-0.9.1-trace.log) - [server-0.9.2-trace.log](https://github.com/user-attachments/files/21064566/server-0.9.2-trace.log) ### OS Windows ### GPU AMD ### CPU AMD ### Ollama version 0.9.2 and up
GiteaMirror added the bug label 2026-04-29 05:01:17 -05:00
GiteaMirror commented 2026-04-29 05:01:20 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 4, 2025):

Logs?

<!-- gh-comment-id:3036991721 --> @rick-github commented on GitHub (Jul 4, 2025): Logs?
GiteaMirror commented 2026-04-29 05:01:22 -05:00
Author
Owner
Copy Link

@Stogas commented on GitHub (Jul 4, 2025):

Sorry, updated description - issue template messed with the formatting.

<!-- gh-comment-id:3037051294 --> @Stogas commented on GitHub (Jul 4, 2025): Sorry, updated description - issue template messed with the formatting.
GiteaMirror commented 2026-04-29 05:01:22 -05:00
Author
Owner
Copy Link

@Stogas commented on GitHub (Jul 4, 2025):

On the main branch, reverting 9e125d884c seems to fix the issue.

<!-- gh-comment-id:3037077266 --> @Stogas commented on GitHub (Jul 4, 2025): On the main branch, reverting https://github.com/ollama/ollama/commit/9e125d884cf995dfae7fcd74690d525e4326a517 seems to fix the issue.
GiteaMirror commented 2026-04-29 05:01:23 -05:00
Author
Owner
Copy Link

@2jfs904judsw20600jikn613d0dookl23jsig commented on GitHub (Jul 5, 2025):

I can't add anything technical here other than that I've noticed this as well. Long GPU inference times with no erroring. Extreme slowdowns with certain model-params.

<!-- gh-comment-id:3038437973 --> @2jfs904judsw20600jikn613d0dookl23jsig commented on GitHub (Jul 5, 2025): I can't add anything technical here other than that I've noticed this as well. Long GPU inference times with no erroring. Extreme slowdowns with certain model-params.
GiteaMirror commented 2026-04-29 05:01:24 -05:00
Author
Owner
Copy Link

@Stogas commented on GitHub (Jul 10, 2025):

@rick-github genuine question - should I keep the PR open without any reviewers, i.e. it'll get a look when someone has time? I'm not familiar with the review process in this repo

<!-- gh-comment-id:3057612245 --> @Stogas commented on GitHub (Jul 10, 2025): @rick-github genuine question - should I keep the PR open without any reviewers, i.e. it'll get a look when someone has time? I'm not familiar with the review process in this repo
GiteaMirror commented 2026-04-29 05:01:24 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 10, 2025):

It will get looked at eventually, but it can take some time - I've had PRs that have taken months to get an initial review. The ollama team prioritize code-touching PRs from the core team, so PRs from external contributors can languish.

<!-- gh-comment-id:3057708481 --> @rick-github commented on GitHub (Jul 10, 2025): It will get looked at eventually, but it can take some time - I've had PRs that have taken months to get an initial review. The ollama team prioritize code-touching PRs from the core team, so PRs from external contributors can languish.
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#53966
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?