github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #8723] VRAM not being fully utilized #31417

New Issue
Closed
opened 2026-04-22 11:51:05 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-22 11:51:05 -05:00
Owner
Copy Link

Originally created by @linh1987 on GitHub (Jan 31, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8723

What is the issue?

Hey guys,
I notice that ollama doesn't make full use of my 3090's VRAM. it seems from the log that the detected VRAM available (~22.7GB) is lower then the real available VRAM reported by nvidia-smi (~23.8GB). The resulted effect is for some model, I can't fit both model and context into VRAM, while I can squeeze the same amount of layers and context cache fully in VRAM using koboldcpp.

Is there anything that we can do to get it to work? I attached the relevant output from nvidia-smi and ollama's server log. I use q8 kv cache quantization.

ollama.txt

OS

Windows

GPU

Nvidia

CPU

AMD

Ollama version

0.5.7

Originally created by @linh1987 on GitHub (Jan 31, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8723 ### What is the issue? Hey guys, I notice that ollama doesn't make full use of my 3090's VRAM. it seems from the log that the detected VRAM available (~22.7GB) is lower then the real available VRAM reported by nvidia-smi (~23.8GB). The resulted effect is for some model, I can't fit both model and context into VRAM, while I can squeeze the same amount of layers and context cache fully in VRAM using koboldcpp. Is there anything that we can do to get it to work? I attached the relevant output from nvidia-smi and ollama's server log. I use q8 kv cache quantization. [ollama.txt](https://github.com/user-attachments/files/18615328/ollama.txt) ### OS Windows ### GPU Nvidia ### CPU AMD ### Ollama version 0.5.7
GiteaMirror added the bug label 2026-04-22 11:51:05 -05:00
GiteaMirror commented 2026-04-22 11:51:07 -05:00
Author
Owner
Copy Link

@qits-kkruse commented on GitHub (Jan 31, 2025):

Do you run your GUI on that same card? I believe it needs a bit of VRAM as well, 1.1GB VRAM looks reasonable to me.

My GUI (KDE/Wayland + Firefox and lots of tabs) uses 1.6GB.

<!-- gh-comment-id:2627544245 --> @qits-kkruse commented on GitHub (Jan 31, 2025): Do you run your GUI on that same card? I believe it needs a bit of VRAM as well, 1.1GB VRAM looks reasonable to me. My GUI (KDE/Wayland + Firefox and lots of tabs) uses 1.6GB.
GiteaMirror commented 2026-04-22 11:51:08 -05:00
Author
Owner
Copy Link

@linh1987 commented on GitHub (Jan 31, 2025):

I run on windows and I use only 0.3GB of VRAM before the first model is run, even then only 22.7GB VRAM is reported to be available. Nothing else is run on the computer.

<!-- gh-comment-id:2627566336 --> @linh1987 commented on GitHub (Jan 31, 2025): I run on windows and I use only 0.3GB of VRAM before the first model is run, even then only 22.7GB VRAM is reported to be available. Nothing else is run on the computer.
GiteaMirror commented 2026-04-22 11:51:08 -05:00
Author
Owner
Copy Link

@tie-pilot-qxw commented on GitHub (Feb 1, 2025):

I have the same issue when running on my 4060 laptop on WSL. There is a total of 8.0GB, but only 6.9 GB is available. I have raised issue #7996, but it's not solved yet.

<!-- gh-comment-id:2628734936 --> @tie-pilot-qxw commented on GitHub (Feb 1, 2025): I have the same issue when running on my 4060 laptop on WSL. There is a total of 8.0GB, but only 6.9 GB is available. I have raised issue #7996, but it's not solved yet.
GiteaMirror commented 2026-04-22 11:51:10 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 5, 2026):

Memory estimation for some models is inaccurate. Recent versions run most models with a new allocation mechanism which improves memory utilisation. I'm going to close this as stale, if you still have problems open a new issue and include a full server log.

<!-- gh-comment-id:3709705709 --> @rick-github commented on GitHub (Jan 5, 2026): Memory estimation for some models is inaccurate. Recent versions run most models with a new allocation mechanism which improves memory utilisation. I'm going to close this as stale, if you still have problems open a new issue and include a full [server log](https://docs.ollama.com/troubleshooting).
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
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#31417
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?