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 14.6k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #6453] Inconsistent GPU Usage #50570

New Issue
Closed
opened 2026-04-28 16:24:59 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-28 16:24:59 -05:00
Owner
Copy Link

Originally created by @gru3zi on GitHub (Aug 21, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6453

What is the issue?

I have been happily using Ollama for sometime with my Dual RTX 3090's with an NV-Link Adaptor. Recently ive been finding the output to be quite slow. After checking both the outputs of 'ollama ps' and nvidia-smi I found that my GPUs are not fully being utilised anymore. It seems to split between the CPU and GPUs.

If I run different models of the same size, some output full GPU usage while others dont.

image

2024-08-21_15-48

My service file which I set with Environment=CUDA_VISIBLE_DEVICES and had also run 'sudo rmmod nvidia_uvm && sudo modprobe nvidia_uvm'

image

Is there a way to stop CPU usage all together?

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

ollama version is 0.3.6

Originally created by @gru3zi on GitHub (Aug 21, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6453 ### What is the issue? I have been happily using Ollama for sometime with my Dual RTX 3090's with an NV-Link Adaptor. Recently ive been finding the output to be quite slow. After checking both the outputs of 'ollama ps' and nvidia-smi I found that my GPUs are not fully being utilised anymore. It seems to split between the CPU and GPUs. If I run different models of the same size, some output full GPU usage while others dont. ![image](https://github.com/user-attachments/assets/4e2746ca-e8dc-4051-935b-5a99c2344600) ![2024-08-21_15-48](https://github.com/user-attachments/assets/d1099028-3146-4292-96ae-47151589f487) My service file which I set with **Environment=CUDA_VISIBLE_DEVICES** and had also run 'sudo rmmod nvidia_uvm && sudo modprobe nvidia_uvm' ![image](https://github.com/user-attachments/assets/9f0254c1-2db2-43e7-b5f1-01e94d1ee528) Is there a way to stop CPU usage all together? ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version ollama version is 0.3.6
GiteaMirror added the bug label 2026-04-28 16:24:59 -05:00
GiteaMirror commented 2026-04-28 16:25:04 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 21, 2024):

Server logs may give insight in to how ollama is scheduling the models. Based on the screenshots it looks like it's only detecting one GPU.

<!-- gh-comment-id:2302361033 --> @rick-github commented on GitHub (Aug 21, 2024): [Server logs](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) may give insight in to how ollama is scheduling the models. Based on the screenshots it looks like it's only detecting one GPU.
GiteaMirror commented 2026-04-28 16:25:05 -05:00
Author
Owner
Copy Link

@gru3zi commented on GitHub (Aug 21, 2024):

I made some changes and now it uses all my GPUs again.

[Unit]
Description=Ollama Service
After=network-online.target

[Service]
ExecStart=/usr/local/bin/ollama serve
User=ollama
Group=ollama
Restart=always
RestartSec=3
Environment="PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:>
CUDA_VISIBLE_DEVICES=0,1
OLLAMA_SCHED_SPREAD=1

[Install]
WantedBy=default.target

<!-- gh-comment-id:2302901676 --> @gru3zi commented on GitHub (Aug 21, 2024): I made some changes and now it uses all my GPUs again. [Unit] Description=Ollama Service After=network-online.target [Service] ExecStart=/usr/local/bin/ollama serve User=ollama Group=ollama Restart=always RestartSec=3 Environment="PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:> **CUDA_VISIBLE_DEVICES=0,1** **OLLAMA_SCHED_SPREAD=1** [Install] WantedBy=default.target
GiteaMirror commented 2026-04-28 16:25:06 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 21, 2024):

If your intent is pass the variables CUDA_VISIBLE_DEVICES and OLLAMA_SCHED_SPREAD to the ollama server, you need to add them as Environment directions.

Environment="CUDA_VISIBLE_DEVICES=0,1"
Environment="OLLAMA_SCHED_SPREAD=1"
<!-- gh-comment-id:2302921872 --> @rick-github commented on GitHub (Aug 21, 2024): If your intent is pass the variables `CUDA_VISIBLE_DEVICES` and `OLLAMA_SCHED_SPREAD` to the ollama server, you need to add them as `Environment` directions. ``` Environment="CUDA_VISIBLE_DEVICES=0,1" Environment="OLLAMA_SCHED_SPREAD=1" ```
GiteaMirror commented 2026-04-28 16:25:07 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Aug 21, 2024):

The problem seems to be the content of CUDA_VISIBLE_DEVICES in your original systemd configuration.

CUDA_VISIBLE_DEVICES (and HIP_VISIBLE_DEVICES) should be comma separated without spaces. Your original line contained a space between the comma and the next device.

Environment="CUDA_VISIBLE_DEVICES=GPU-986c10fb-51e6-ba79-67dd-fd9e95d31034, GPU-d742ac89-cd44-c238>

Environment="CUDA_VISIBLE_DEVICES=GPU-986c10fb-51e6-ba79-67dd-fd9e95d31034,GPU-d742ac89-cd44-c238>
<!-- gh-comment-id:2302932917 --> @mxyng commented on GitHub (Aug 21, 2024): The problem seems to be the content of `CUDA_VISIBLE_DEVICES` in your original systemd configuration. `CUDA_VISIBLE_DEVICES` (and `HIP_VISIBLE_DEVICES`) should be comma separated _without_ spaces. Your original line contained a space between the comma and the next device. ``` Environment="CUDA_VISIBLE_DEVICES=GPU-986c10fb-51e6-ba79-67dd-fd9e95d31034, GPU-d742ac89-cd44-c238> ``` ``` Environment="CUDA_VISIBLE_DEVICES=GPU-986c10fb-51e6-ba79-67dd-fd9e95d31034,GPU-d742ac89-cd44-c238> ```
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#50570
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?