github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-21 13:11:39 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #6091] Parallel Bug: Would rather queue than reload on another GPU #50319

New Issue
Closed
opened 2026-04-28 15:06:10 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-28 15:06:10 -05:00
Owner
Copy Link

Originally created by @txd0213 on GitHub (Jul 31, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6091

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

Experimental environment: 8 x A6000 GPUs
LLM: qwen2:7b
Environment variables:

Environment="OLLAMA_NUM_PARALLEL=16"
Environment="OLLAMA_MAX_LOADED_MODELS=4"

When the concurrency is less than or equal to 4, the parallel processing is effective. However, once it exceeds 4, OLLAMA does not choose to reload the same model on another GPU.

image

Although I sent 16 requests simultaneously, as can be seen from the graph, the actual concurrency of the model is only 4.
image

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.3.0

Originally created by @txd0213 on GitHub (Jul 31, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6091 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? **Experimental environment: 8 x A6000 GPUs** **LLM: qwen2:7b** **Environment variables:** ``` Environment="OLLAMA_NUM_PARALLEL=16" Environment="OLLAMA_MAX_LOADED_MODELS=4" ``` When the concurrency is less than or equal to **4**, the parallel processing is effective. However, once it exceeds 4, OLLAMA does not choose to **reload the same model on another GPU**. ![image](https://github.com/user-attachments/assets/de471e9e-1e21-4e5c-969f-8a17d0a0db19) Although I sent 16 requests simultaneously, as can be seen from the graph, the actual concurrency of the model is only 4. ![image](https://github.com/user-attachments/assets/0e0893c0-84c5-47cf-9bdb-528c1b24cd92) ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.3.0
GiteaMirror added the feature request label 2026-04-28 15:06:10 -05:00
GiteaMirror commented 2026-04-28 15:06:15 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Aug 1, 2024):

I think what you're asking for is the ability to load the same model multiple times on different GPUs. That's not currently supported with our concurrency implementation, and tracked via #3902 so I'll close this as a dup.

If that's not what your asking, can you clarify and I'll reopen the issue.

<!-- gh-comment-id:2264111911 --> @dhiltgen commented on GitHub (Aug 1, 2024): I think what you're asking for is the ability to load the same model multiple times on different GPUs. That's not currently supported with our concurrency implementation, and tracked via #3902 so I'll close this as a dup. If that's not what your asking, can you clarify and I'll reopen the issue.
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 feature request
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#50319
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?