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 #2156] I want to run Ollama on the limited number of GPUS and CPUS #63269

New Issue
Closed
opened 2026-05-03 12:47:34 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-03 12:47:34 -05:00
Owner
Copy Link

Originally created by @sfarzi on GitHub (Jan 23, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/2156

I have a machine with 4 GPUS and 16 CPUS. but I want to run Ollama just on one gpu and 8 cpus. How can I do this?

Originally created by @sfarzi on GitHub (Jan 23, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/2156 I have a machine with 4 GPUS and 16 CPUS. but I want to run Ollama just on one gpu and 8 cpus. How can I do this?
GiteaMirror commented 2026-05-03 12:47:35 -05:00
Author
Owner
Copy Link

@easp commented on GitHub (Jan 23, 2024):

Manuall setting num_thread in a Modelfile will limit the cores used. GPUs is more complicated. I think this will work (assuming you are using NVIDIA)
https://developer.nvidia.com/blog/cuda-pro-tip-control-gpu-visibility-cuda_visible_devices/

<!-- gh-comment-id:1906520931 --> @easp commented on GitHub (Jan 23, 2024): Manuall setting num_thread in a Modelfile will limit the cores used. GPUs is more complicated. I think this will work (assuming you are using NVIDIA) https://developer.nvidia.com/blog/cuda-pro-tip-control-gpu-visibility-cuda_visible_devices/
GiteaMirror commented 2026-05-03 12:47:35 -05:00
Author
Owner
Copy Link

@jukofyork commented on GitHub (Jan 24, 2024):

There is already the option to pass through the main_gpu option to the wrapped llama.cpp server but the patch to pass through the tensor_split option https://github.com/ollama/ollama/pull/1256 seems to be stuck and says: "This branch has conflicts that must be resolved".

Somebody in that thread replied that the patch works fine though.

<!-- gh-comment-id:1907918774 --> @jukofyork commented on GitHub (Jan 24, 2024): There is already the option to pass through the `main_gpu` option to the wrapped llama.cpp server but the patch to pass through the `tensor_split` option https://github.com/ollama/ollama/pull/1256 seems to be stuck and says: "This branch has conflicts that must be resolved". Somebody in that thread replied that the patch works fine though.
GiteaMirror commented 2026-05-03 12:47:36 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Mar 12, 2024):

I've updated our docs to show how to limit GPUs.

You can experiment with taskset for CPU limiting, but I would suggest using something like nice to adjust the priority of the process if you're trying to avoid starving other workloads of CPU cycles.

<!-- gh-comment-id:1992311810 --> @dhiltgen commented on GitHub (Mar 12, 2024): I've updated our docs to show how to limit GPUs. You can experiment with `taskset` for CPU limiting, but I would suggest using something like `nice` to adjust the priority of the process if you're trying to avoid starving other workloads of CPU cycles.
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#63269
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?