github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 00:51:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #4668] Low GPU / High CPU Utilization ==> Slow Performance #2935

New Issue
Closed
opened 2026-04-12 13:18:25 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-12 13:18:25 -05:00
Owner
Copy Link

Originally created by @tarekeldeeb on GitHub (May 27, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4668

What is the issue?

The Ollama on Ubuntu 22.04 can detect my Cuda GPU, and loads the model to its memory, but the processing seems to be mostly on CPU. Is this a normal behavior? The overall performance is not satisfying, like 1-token-per-second or so ... much slower than a human reading speed.
Screenshot from 2024-05-27 15-04-47

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.1.38

Originally created by @tarekeldeeb on GitHub (May 27, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4668 ### What is the issue? The Ollama on Ubuntu 22.04 can detect my Cuda GPU, and loads the model to its memory, but the processing seems to be mostly on CPU. Is this a normal behavior? The overall performance is not satisfying, like 1-token-per-second or so ... much slower than a human reading speed. ![Screenshot from 2024-05-27 15-04-47](https://github.com/ollama/ollama/assets/90985/bc949a87-ced5-45f8-bea3-a99dfc739e19) ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.1.38
GiteaMirror added the bug label 2026-04-12 13:18:25 -05:00
GiteaMirror commented 2026-04-12 13:18:27 -05:00
Author
Owner
Copy Link

@easp commented on GitHub (May 27, 2024):

What model and quantization?

If the model doesn't fit entirely on the GPU, part of it will run out of system memory on the CPU. GPU utilization will be low because GPU has to wait for the slow CPU.

More insight is available through ollama ps

<!-- gh-comment-id:2133884799 --> @easp commented on GitHub (May 27, 2024): What model and quantization? If the model doesn't fit entirely on the GPU, part of it will run out of system memory on the CPU. GPU utilization will be low because GPU has to wait for the slow CPU. More insight is available through `ollama ps`
GiteaMirror commented 2026-04-12 13:18:27 -05:00
Author
Owner
Copy Link

@tarekeldeeb commented on GitHub (May 28, 2024):

Makes sense now, the model does not totally fit in GPU!

ollama ps
NAME        	ID          	SIZE  	PROCESSOR      	UNTIL               
codellama:7b	8fdf8f752f6e	6.8 GB	43%/57% CPU/GPU	25 minutes from now
<!-- gh-comment-id:2134477784 --> @tarekeldeeb commented on GitHub (May 28, 2024): Makes sense now, the model does not totally fit in GPU! <pre>ollama ps NAME ID SIZE PROCESSOR UNTIL codellama:7b 8fdf8f752f6e 6.8 GB 43%/57% CPU/GPU 25 minutes from now</pre>
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#2935
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?