github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 17:49:24 -05:00
Code Issues 14.5k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #13860] Serious performance regression on ARM64 systems with 0.14.x ollama version #55586

New Issue
Closed
opened 2026-04-29 09:27:13 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-29 09:27:13 -05:00
Owner
Copy Link

Originally created by @rjtokenring on GitHub (Jan 23, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/13860

What is the issue?

Hi, I'm using ollama on ARM64 . So in CPU mode.
I noticed a serious performance drop in upgrading to version 0.14.3.
Throughput dropped by 10x.

Using ollama 0.12.1 and 0.14.3 - ARM64 container version.

Host: Ubuntu 24.04

Relevant log output

With
docker run -it -v /home/ubuntu/models/:/root/.ollama -p 11434:11434 ollama/ollama:0.14.3

---------------------------------
>>> Hi!
Hi there! How’s your day going? 😊

What can I help you with today?

total duration:       32.759942325s
load duration:        541.75404ms
prompt eval count:    11 token(s)
prompt eval duration: 8.373959573s
prompt eval rate:     1.31 tokens/s
eval count:           22 token(s)
eval duration:        23.722800925s
eval rate:            0.93 tokens/s



With
docker run -it -v /home/ubuntu/models/:/root/.ollama -p 11434:11434 ollama/ollama:0.12.1

---------------------------------

>>> Hi!
Hi there! How's your day going so far? 😊

Is there anything you'd like to chat about or any help I can offer?

total duration:       4.066736884s
load duration:        431.213999ms
prompt eval count:    11 token(s)
prompt eval duration: 320.176938ms
prompt eval rate:     34.36 tokens/s
eval count:           33 token(s)
eval duration:        3.165976348s
eval rate:            10.42 tokens/s

OS

Docker

GPU

No response

CPU

Other

Ollama version

0.14.3

Originally created by @rjtokenring on GitHub (Jan 23, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/13860 ### What is the issue? Hi, I'm using ollama on ARM64 . So in CPU mode. I noticed a serious performance drop in upgrading to version 0.14.3. Throughput dropped by 10x. Using ollama 0.12.1 and 0.14.3 - ARM64 container version. Host: Ubuntu 24.04 ### Relevant log output ```shell With docker run -it -v /home/ubuntu/models/:/root/.ollama -p 11434:11434 ollama/ollama:0.14.3 --------------------------------- >>> Hi! Hi there! How’s your day going? 😊 What can I help you with today? total duration: 32.759942325s load duration: 541.75404ms prompt eval count: 11 token(s) prompt eval duration: 8.373959573s prompt eval rate: 1.31 tokens/s eval count: 22 token(s) eval duration: 23.722800925s eval rate: 0.93 tokens/s With docker run -it -v /home/ubuntu/models/:/root/.ollama -p 11434:11434 ollama/ollama:0.12.1 --------------------------------- >>> Hi! Hi there! How's your day going so far? 😊 Is there anything you'd like to chat about or any help I can offer? total duration: 4.066736884s load duration: 431.213999ms prompt eval count: 11 token(s) prompt eval duration: 320.176938ms prompt eval rate: 34.36 tokens/s eval count: 33 token(s) eval duration: 3.165976348s eval rate: 10.42 tokens/s ``` ### OS Docker ### GPU _No response_ ### CPU Other ### Ollama version 0.14.3
GiteaMirror added the bug label 2026-04-29 09:27:13 -05:00
GiteaMirror commented 2026-04-29 09:27:14 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 23, 2026):

Confirm performance issue with 0.14.2->0.14.3, continues in 0.15.0-rc1.

0.14.2     prompt: 369.71 eval: 20.94 prompt: 389.94 eval: 22.09 prompt: 639.82 eval: 22.72 prompt: 704.21 eval: 24.85 prompt: 526.89 eval: 20.96
0.14.3     prompt: 19.56 eval: 7.12 prompt: 209.67 eval: 7.20 prompt: 330.19 eval: 7.30 prompt: 235.31 eval: 7.31 prompt: 330.94 eval: 7.22
0.15.0-rc1 prompt: 19.92 eval: 7.48 prompt: 238.43 eval: 7.32 prompt: 330.77 eval: 7.26 prompt: 330.56 eval: 7.37 prompt: 229.11 eval: 7.20
<!-- gh-comment-id:3790811653 --> @rick-github commented on GitHub (Jan 23, 2026): Confirm performance issue with 0.14.2->0.14.3, continues in 0.15.0-rc1. ``` 0.14.2 prompt: 369.71 eval: 20.94 prompt: 389.94 eval: 22.09 prompt: 639.82 eval: 22.72 prompt: 704.21 eval: 24.85 prompt: 526.89 eval: 20.96 0.14.3 prompt: 19.56 eval: 7.12 prompt: 209.67 eval: 7.20 prompt: 330.19 eval: 7.30 prompt: 235.31 eval: 7.31 prompt: 330.94 eval: 7.22 0.15.0-rc1 prompt: 19.92 eval: 7.48 prompt: 238.43 eval: 7.32 prompt: 330.77 eval: 7.26 prompt: 330.56 eval: 7.37 prompt: 229.11 eval: 7.20 ```
GiteaMirror commented 2026-04-29 09:27:14 -05:00
Author
Owner
Copy Link

@mchiang0610 commented on GitHub (Jan 24, 2026):

Thanks @rjtokenring filing this and @rick-github for reproducing. May I ask what hardware you are using? We should add something similar to our hardware testing cluster.

<!-- gh-comment-id:3793749273 --> @mchiang0610 commented on GitHub (Jan 24, 2026): Thanks @rjtokenring filing this and @rick-github for reproducing. May I ask what hardware you are using? We should add something similar to our hardware testing cluster.
GiteaMirror commented 2026-04-29 09:27:15 -05:00
Author
Owner
Copy Link

@rjtokenring commented on GitHub (Jan 24, 2026):

Reproduced on Qualcomm 6490, 8275 and 2210.
Recompiling main on debian trixie seems to fix the problem.

<!-- gh-comment-id:3793996912 --> @rjtokenring commented on GitHub (Jan 24, 2026): Reproduced on Qualcomm 6490, 8275 and 2210. Recompiling main on debian trixie seems to fix the problem.
GiteaMirror commented 2026-04-29 09:27:15 -05:00
Author
Owner
Copy Link

@crumble0815 commented on GitHub (Jan 24, 2026):

Reproduced on Raspberry Pi 5 16GB.
Trixie with stable and experimental kernel 6.18.6-v8-16k+
100% CPU with model phi4

Windows on Arm (SQ2) feels only a little slower than 13.x

I assume that linux version does not use NEON

added stderr from ollama serve
performanceIssuePi5.log

<!-- gh-comment-id:3794773204 --> @crumble0815 commented on GitHub (Jan 24, 2026): Reproduced on Raspberry Pi 5 16GB. Trixie with stable and experimental kernel 6.18.6-v8-16k+ 100% CPU with model phi4 Windows on Arm (SQ2) feels only a little slower than 13.x I assume that linux version does not use NEON added stderr from ollama serve [performanceIssuePi5.log](https://github.com/user-attachments/files/24837781/performanceIssuePi5.log)
GiteaMirror commented 2026-04-29 09:27:16 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 25, 2026):

0.15.0-rc1 prompt: 19.79 eval: 7.02 prompt: 333.01 eval: 7.16 prompt: 292.77 eval: 7.17 prompt: 333.07 eval: 7.10 prompt: 330.62 eval: 7.49
0.15.0     prompt: 19.80 eval: 7.30 prompt: 332.39 eval: 7.24 prompt: 233.94 eval: 7.72 prompt: 286.12 eval: 7.40 prompt: 331.43 eval: 6.87
0.15.1     prompt: 561.51 eval: 23.02 prompt: 5239.68 eval: 20.08 prompt: 596.83 eval: 22.91 prompt: 432.20 eval: 25.98 prompt: 5186.53 eval: 27.94
<!-- gh-comment-id:3797086443 --> @rick-github commented on GitHub (Jan 25, 2026): ``` 0.15.0-rc1 prompt: 19.79 eval: 7.02 prompt: 333.01 eval: 7.16 prompt: 292.77 eval: 7.17 prompt: 333.07 eval: 7.10 prompt: 330.62 eval: 7.49 0.15.0 prompt: 19.80 eval: 7.30 prompt: 332.39 eval: 7.24 prompt: 233.94 eval: 7.72 prompt: 286.12 eval: 7.40 prompt: 331.43 eval: 6.87 0.15.1 prompt: 561.51 eval: 23.02 prompt: 5239.68 eval: 20.08 prompt: 596.83 eval: 22.91 prompt: 432.20 eval: 25.98 prompt: 5186.53 eval: 27.94 ```
GiteaMirror commented 2026-04-29 09:27:16 -05:00
Author
Owner
Copy Link

@rjtokenring commented on GitHub (Jan 26, 2026):

@mchiang0610 @rick-github tested 0.15.1 and I can confirm that regression has been fixed!

Thanks!

<!-- gh-comment-id:3800739024 --> @rjtokenring commented on GitHub (Jan 26, 2026): @mchiang0610 @rick-github tested 0.15.1 and I can confirm that regression has been fixed! Thanks!
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#55586
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?