github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #6420] Is the speed of the Olama running model related to the CUDA version? #66074

New Issue
Closed
opened 2026-05-03 23:51:39 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-03 23:51:39 -05:00
Owner
Copy Link

Originally created by @TianWuYuJiangHenShou on GitHub (Aug 19, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6420

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

I deployed qwen2:72B with the latest version of Ollama, but I found that the loading speed of Ollama models varies greatly under different nvidia driver version.

driver:535.183.06 | cuda version:12.2

ollama version:0.3.4
Time of Loading Model:29s

driver:515.105.01 | cuda version:11.7

ollama version:0.3.6
Time of Loading Model:659s

GPU :A800

OS

Linux

GPU

Nvidia

CPU

No response

Ollama version

No response

Originally created by @TianWuYuJiangHenShou on GitHub (Aug 19, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6420 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? I deployed qwen2:72B with the latest version of Ollama, but I found that the loading speed of Ollama models varies greatly under different nvidia driver version. # driver:535.183.06 | cuda version:12.2 ollama version:0.3.4 Time of Loading Model:29s # driver:515.105.01 | cuda version:11.7 ollama version:0.3.6 Time of Loading Model:659s ## GPU :A800 ### OS Linux ### GPU Nvidia ### CPU _No response_ ### Ollama version _No response_
GiteaMirror added the gpubugnvidianeeds more info labels 2026-05-03 23:51:40 -05:00
GiteaMirror commented 2026-05-03 23:51:44 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 19, 2024):

If the model is already in the buffer cache, reloading will be a lot faster. Try tthis:

time curl localhost:11434/api/generate -d '{"model":"qwen2:72b","prompt":"hi","options":{"seed":0},"stream":false,"keep_alive":0}'
sleep 5
time curl localhost:11434/api/generate -d '{"model":"qwen2:72b","prompt":"hi","options":{"seed":0},"stream":false,"keep_alive":0}'

The second command will load the model from buffer cache and will give you a better measure of how long it takes the cuda driver to load the model into the GPU.

<!-- gh-comment-id:2296288197 --> @rick-github commented on GitHub (Aug 19, 2024): If the model is already in the buffer cache, reloading will be a lot faster. Try tthis: ``` time curl localhost:11434/api/generate -d '{"model":"qwen2:72b","prompt":"hi","options":{"seed":0},"stream":false,"keep_alive":0}' sleep 5 time curl localhost:11434/api/generate -d '{"model":"qwen2:72b","prompt":"hi","options":{"seed":0},"stream":false,"keep_alive":0}' ``` The second command will load the model from buffer cache and will give you a better measure of how long it takes the cuda driver to load the model into the GPU.
GiteaMirror commented 2026-05-03 23:51:46 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Sep 3, 2024):

I wouldn't expect the driver version alone to make that large of a difference.

Can you share your server log for both scenarios so we can see more details?

<!-- gh-comment-id:2327615605 --> @dhiltgen commented on GitHub (Sep 3, 2024): I wouldn't expect the driver version alone to make that large of a difference. Can you share your server log for both scenarios so we can see more details?
GiteaMirror commented 2026-05-03 23:51:47 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Sep 26, 2024):

If you're still seeing performance problems, please share your server log and I'll reopen the issue.

<!-- gh-comment-id:2375485835 --> @dhiltgen commented on GitHub (Sep 26, 2024): If you're still seeing performance problems, please share your server log 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 bug gpu needs more info nvidia
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#66074
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?