github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 05:05:37 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #9753] Ollama's output speed is too slow compared to the GPU specifications. #52884

New Issue
Closed
opened 2026-04-29 01:16:15 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-29 01:16:15 -05:00
Owner
Copy Link

Originally created by @udkii on GitHub (Mar 14, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9753

What is the issue?

When I load a model from Ollama and run it on my computer, the GPU does not increase, but only the VRAM increases. When I ask a question, the GPU usage increases for a moment, so it seems that the calculation is finished at that moment. However, after that, the GPU usage is always around 2%, and only the VRAM and RAM memory are full, so it takes 10 minutes to generate the answer. Did I set this up wrong when I used Ollama? Is the GPU not being used right now? I did the basic torch.cuda.is_available(), and it said True. I wonder if there is a setting I missed.

My computer specifications are like this
CPU : Intel i9-10920X 3.5GHZ
RAM : 32GB * 2
GPU: NVIDIA RTX A5000 24GB
model : llama 3.3 70B-q4_K_M

Image

Relevant log output

OS

No response

GPU

No response

CPU

No response

Ollama version

No response

Originally created by @udkii on GitHub (Mar 14, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9753 ### What is the issue? When I load a model from Ollama and run it on my computer, the GPU does not increase, but only the VRAM increases. When I ask a question, the GPU usage increases for a moment, so it seems that the calculation is finished at that moment. However, after that, the GPU usage is always around 2%, and only the VRAM and RAM memory are full, so it takes 10 minutes to generate the answer. Did I set this up wrong when I used Ollama? Is the GPU not being used right now? I did the basic torch.cuda.is_available(), and it said True. I wonder if there is a setting I missed. My computer specifications are like this CPU : Intel i9-10920X 3.5GHZ RAM : 32GB * 2 GPU: NVIDIA RTX A5000 24GB model : llama 3.3 70B-q4_K_M ![Image](https://github.com/user-attachments/assets/7844de49-321a-4a1f-a441-39c6e0fd56f4) ### Relevant log output ```shell ``` ### OS _No response_ ### GPU _No response_ ### CPU _No response_ ### Ollama version _No response_
GiteaMirror added the bugneeds more info labels 2026-04-29 01:16:15 -05:00
GiteaMirror commented 2026-04-29 01:16:19 -05:00
Author
Owner
Copy Link

@MMaturax commented on GitHub (Mar 14, 2025):

for LLaMA 3.3 70B-Q4_K_M, you need 53GB of VRAM. Since your GPU only has 24GB, it's also using system RAM, and as a result, you're likely relying more on CPU than GPU. If you want to achieve optimal performance, I suggest trying smaller models.

<!-- gh-comment-id:2723734544 --> @MMaturax commented on GitHub (Mar 14, 2025): for LLaMA 3.3 70B-Q4_K_M, you need 53GB of VRAM. Since your GPU only has 24GB, it's also using system RAM, and as a result, you're likely relying more on CPU than GPU. If you want to achieve optimal performance, I suggest trying smaller models.
GiteaMirror commented 2026-04-29 01:16:20 -05:00
Author
Owner
Copy Link

@mbretter commented on GitHub (Mar 14, 2025):

run ollama ps, the output should show something like:
llama3.1:latest 46e0c10c039e 6.9 GB 100% GPU 4 minutes from now

If you do not have enough GRAM, you will get something like this:
llama3.2:latest a80c4f17acd5 3.4 GB 10%/90% CPU/GPU 4 minutes from now

And performance will drop down.

<!-- gh-comment-id:2723801549 --> @mbretter commented on GitHub (Mar 14, 2025): run `ollama ps`, the output should show something like: `llama3.1:latest 46e0c10c039e 6.9 GB 100% GPU 4 minutes from now ` If you do not have enough GRAM, you will get something like this: `llama3.2:latest a80c4f17acd5 3.4 GB 10%/90% CPU/GPU 4 minutes from now ` And performance will drop down.
GiteaMirror commented 2026-04-29 01:16:21 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Mar 19, 2025):

@udkii can you post the output of ollama ps?

<!-- gh-comment-id:2735349117 --> @pdevine commented on GitHub (Mar 19, 2025): @udkii can you post the output of `ollama ps`?
GiteaMirror commented 2026-04-29 01:16:21 -05:00
Author
Owner
Copy Link

@udkii commented on GitHub (Mar 26, 2025):

@udkii can you post the output of ollama ps?

@pdevine Sorry for the late reply. Is it really because my VRAM size is small? My ps is as follows. It's using CPU 70%.

Image

<!-- gh-comment-id:2753045239 --> @udkii commented on GitHub (Mar 26, 2025): > [@udkii](https://github.com/udkii) can you post the output of `ollama ps`? @pdevine Sorry for the late reply. Is it really because my VRAM size is small? My ps is as follows. It's using CPU 70%. ![Image](https://github.com/user-attachments/assets/6c3fa567-8f53-4bf0-b993-d6dac8c1011f)
GiteaMirror commented 2026-04-29 01:16:22 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Mar 26, 2025):

@udkii yes, you're running primarily on your CPU because you don't have enough VRAM for the model. Any time you split the model like that on the CPU/GPU it's going to go a lot slower. Llama3.3 is a really big model, so you might want to consider one of the other Llama3 variants like 3.2 or 3.1.

I'm going to go ahead and close the issue because Ollama is working as expected.

<!-- gh-comment-id:2755169200 --> @pdevine commented on GitHub (Mar 26, 2025): @udkii yes, you're running primarily on your CPU because you don't have enough VRAM for the model. Any time you split the model like that on the CPU/GPU it's going to go a lot slower. Llama3.3 is a really big model, so you might want to consider one of the other Llama3 variants like 3.2 or 3.1. I'm going to go ahead and close the issue because Ollama is working as expected.
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 needs more info
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#52884
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?