github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-22 05:31:43 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #10862] Ollama Multi-threading on aarch64 CPU compared to Llama.cpp #84821

New Issue
Open
opened 2026-05-09 21:49:21 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-05-09 21:49:21 -05:00
Owner
Copy Link

Originally created by @luentong on GitHub (May 26, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10862

I run DeepSeek-R1-Distill-Qwen-7B, Float16 inference using both Ollama and Llama.cpp, and my machine has 24 physical cores.

Ollama creates 37 threads, while 24 of them has non-0% CPU usage, but mostly around 15% - 10% CPU usage.

Image

Llama.cpp creates 25 threads, while 24 of them has close to 100% CPU usage, the 25th thread is the main thread waiting for the other 24 threads, presumably.

Image

Although the CPU usage stats differ significantly, the inference performance remain similar. For Ollama, it's around 9 tokens/s. For Llama.cpp, it's around 9-10 tokens/s.

Anyone knows what kind of abstractions Ollama put on Llama.cpp to make this happen? Thank you

Originally created by @luentong on GitHub (May 26, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10862 I run DeepSeek-R1-Distill-Qwen-7B, Float16 inference using both Ollama and Llama.cpp, and my machine has 24 physical cores. Ollama creates 37 threads, while 24 of them has non-0% CPU usage, but mostly around 15% - 10% CPU usage. ![Image](https://github.com/user-attachments/assets/603695e0-734e-4d92-95a4-ca3e3d764723) Llama.cpp creates 25 threads, while 24 of them has close to 100% CPU usage, the 25th thread is the main thread waiting for the other 24 threads, presumably. ![Image](https://github.com/user-attachments/assets/21e43bdc-31cb-4dc5-8ea6-92c9f4128cf0) Although the CPU usage stats differ significantly, the inference performance remain similar. For Ollama, it's around 9 tokens/s. For Llama.cpp, it's around 9-10 tokens/s. Anyone knows what kind of abstractions Ollama put on Llama.cpp to make this happen? Thank you
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
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#84821
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?