github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #9637] Assign Different Large Models to Each GPU #68344

New Issue
Closed
opened 2026-05-04 13:16:24 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-05-04 13:16:24 -05:00
Owner
Copy Link

Originally created by @ZimaBlueee on GitHub (Mar 11, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9637

Hello,
I have a server with two GPUs, and I would like to run different large models on each GPU. For instance, I want the first GPU to run the Qwen model and the second GPU to run the Llama model. Could you please provide guidance on how to specify which GPU runs which model?
Thank you!

Originally created by @ZimaBlueee on GitHub (Mar 11, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9637 Hello, I have a server with two GPUs, and I would like to run different large models on each GPU. For instance, I want the first GPU to run the Qwen model and the second GPU to run the Llama model. Could you please provide guidance on how to specify which GPU runs which model? Thank you!
GiteaMirror commented 2026-05-04 13:16:29 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 11, 2025):

Just load the models. ollama will try to fit a model on a single GPU, so if qwen takes 60% of a GPU and llama takes 80%, ollama will load one model per GPU. If you are finding that both models fit on a single GPU and you want to spread them out, increase num_ctx on one of the models until the models cannot co-reside.

The other way to do this is to run multiple ollama servers and use CUDA_VISIBLE_DEVICES to bind a server to a GPU.

See #3902 for a placeholder ticket for future work on instance management.

<!-- gh-comment-id:2712245201 --> @rick-github commented on GitHub (Mar 11, 2025): Just load the models. ollama will try to fit a model on a single GPU, so if qwen takes 60% of a GPU and llama takes 80%, ollama will load one model per GPU. If you are finding that both models fit on a single GPU and you want to spread them out, increase `num_ctx` on one of the models until the models cannot co-reside. The other way to do this is to run multiple ollama servers and use `CUDA_VISIBLE_DEVICES` to bind a server to a GPU. See #3902 for a placeholder ticket for future work on instance management.
GiteaMirror commented 2026-05-04 13:16:29 -05:00
Author
Owner
Copy Link

@ZimaBlueee commented on GitHub (Mar 11, 2025):

@rick-github thanks！！

<!-- gh-comment-id:2712696625 --> @ZimaBlueee commented on GitHub (Mar 11, 2025): @rick-github 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
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#68344
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?