github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1666] Ollama not using GPU in Windows WSL2 with Docker #936

New Issue
Closed
opened 2026-04-12 10:38:17 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-12 10:38:17 -05:00
Owner
Copy Link

Originally created by @dominickp on GitHub (Dec 22, 2023).
Original GitHub issue: https://github.com/ollama/ollama/issues/1666

I'm seeing a lot of CPU usage when the model runs. I do see a tiny bit of GPU usage but I don't think what I'm seeing is optimal. I also see log messages saying the GPU is not working.

I'm running Docker Desktop on Windows 11 with WSL2 backend on Ubuntu 22.04.03 LTS. I believe I have the correct drivers installed in Ubuntu.

In the ollama logs:

ollama  | 2023/12/22 00:17:24 routes.go:915: warning: gpu support may not be enabled, check that you have installed GPU drivers: nvidia-smi command failed
...
ollama  | 2023/12/22 00:13:33 llama.go:407: skipping accelerated runner because num_gpu=0
...
ollama  | {"timestamp":1703204013,"level":"WARNING","function":"server_params_parse","line":2160,"message":"Not compiled with GPU offload support, --n-gpu-layers option will be ignored. See main README.md for information on enabling GPU BLAS support","n_gpu_layers":-1}

Inside my WSL2 instance shell:

dom@Dom-14700K:~$ nvidia-smi
Thu Dec 21 19:16:55 2023
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 545.29.04              Driver Version: 546.17       CUDA Version: 12.3     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce RTX 3060        On  | 00000000:01:00.0  On |                  N/A |
|  0%   25C    P5              39W / 170W |    837MiB / 12288MiB |      2%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |

Inside the ollama container itself:

docker-compose exec ollama bash

root@e70cdd37fb90:/# nvidia-smi 
bash: nvidia-smi: command not found
root@e70cdd37fb90:/#

The docker compose I'm using:

version: '3.6'

services:
  ollama:
    # Uncomment below for GPU support
    deploy: {resources: {reservations: {devices: [{driver: nvidia, capabilities: [gpu, video]}]}}}
    volumes:
      - \\wsl$\Ubuntu-22.04\home\dom\ollama:/root/.ollama
    container_name: ollama
    tty: true
    restart: unless-stopped
    image: ollama/ollama:0.1.17

# ... redacted the ollama-webui config

volumes:
  ollama: {}

Am I doing something wrong? I really thought that https://github.com/jmorganca/ollama/pull/1644 would solve my issue. But I'm unsure now because of the recent update implicating Swarm mode, which I think I am not using because I am using Docker Desktop which doesn't support that. I would appreciate any help.

If ollama itself needs to execute nvidia-smi, then shouldn't the container have it installed? I don't understand how the container would be able to reach out onto the Docker host to run nvidia-smi there...

Originally created by @dominickp on GitHub (Dec 22, 2023). Original GitHub issue: https://github.com/ollama/ollama/issues/1666 I'm seeing a lot of CPU usage when the model runs. I do see a tiny bit of GPU usage but I don't think what I'm seeing is optimal. I also see log messages saying the GPU is not working. I'm running Docker Desktop on Windows 11 with WSL2 backend on Ubuntu 22.04.03 LTS. I believe I have the correct drivers installed in Ubuntu. In the ollama logs: ``` ollama | 2023/12/22 00:17:24 routes.go:915: warning: gpu support may not be enabled, check that you have installed GPU drivers: nvidia-smi command failed ... ollama | 2023/12/22 00:13:33 llama.go:407: skipping accelerated runner because num_gpu=0 ... ollama | {"timestamp":1703204013,"level":"WARNING","function":"server_params_parse","line":2160,"message":"Not compiled with GPU offload support, --n-gpu-layers option will be ignored. See main README.md for information on enabling GPU BLAS support","n_gpu_layers":-1} ``` Inside my WSL2 instance shell: ``` dom@Dom-14700K:~$ nvidia-smi Thu Dec 21 19:16:55 2023 +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 545.29.04 Driver Version: 546.17 CUDA Version: 12.3 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 25C P5 39W / 170W | 837MiB / 12288MiB | 2% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | No running processes found | ``` Inside the ollama container itself: ``` docker-compose exec ollama bash root@e70cdd37fb90:/# nvidia-smi bash: nvidia-smi: command not found root@e70cdd37fb90:/# ``` The docker compose I'm using: ```yml version: '3.6' services: ollama: # Uncomment below for GPU support deploy: {resources: {reservations: {devices: [{driver: nvidia, capabilities: [gpu, video]}]}}} volumes: - \\wsl$\Ubuntu-22.04\home\dom\ollama:/root/.ollama container_name: ollama tty: true restart: unless-stopped image: ollama/ollama:0.1.17 # ... redacted the ollama-webui config volumes: ollama: {} ``` Am I doing something wrong? I really thought that https://github.com/jmorganca/ollama/pull/1644 would solve my issue. But I'm unsure now because of the recent update implicating Swarm mode, which I think I am not using because I am using Docker Desktop which doesn't support that. I would appreciate any help. If ollama itself needs to execute `nvidia-smi`, then shouldn't the container have it installed? I don't understand how the container would be able to reach out onto the Docker host to run `nvidia-smi` there...
GiteaMirror commented 2026-04-12 10:38:18 -05:00
Author
Owner
Copy Link

@nuqz1 commented on GitHub (Dec 22, 2023):

i noticed that as well. It seems that the stream itself comes from the GPU but my guess some kind of tokenization/detokenization before that is done on the CPU?

<!-- gh-comment-id:1867089618 --> @nuqz1 commented on GitHub (Dec 22, 2023): i noticed that as well. It seems that the stream itself comes from the GPU but my guess some kind of tokenization/detokenization before that is done on the CPU?
GiteaMirror commented 2026-04-12 10:38:18 -05:00
Author
Owner
Copy Link

@dominickp commented on GitHub (Dec 22, 2023):

I think I figured it out, it's kind of dumb. It seems that providing the "video" capability was breaking everything.

I changed:

services:
  ollama:
    deploy: {resources: {reservations: {devices: [{driver: nvidia, capabilities: [gpu, video]}]}}}
    ...

to:

services:
  ollama:
    deploy: {resources: {reservations: {devices: [{driver: nvidia, count: 1, capabilities: [gpu]}]}}}
    ...

and all the errors went away.

llama2 is super fast now. dolphin-mixtral still takes a long time and uses a ton of CPU -- I wonder if that's because I don't have enough VRAM to run it optimally. But since the errors are gone and I see an improvement, I think I resolved the issue.

<!-- gh-comment-id:1867090712 --> @dominickp commented on GitHub (Dec 22, 2023): I think I figured it out, it's kind of dumb. It seems that providing the "video" capability was breaking everything. I changed: ```yml services: ollama: deploy: {resources: {reservations: {devices: [{driver: nvidia, capabilities: [gpu, video]}]}}} ... ``` to: ```yml services: ollama: deploy: {resources: {reservations: {devices: [{driver: nvidia, count: 1, capabilities: [gpu]}]}}} ... ``` and all the errors went away. llama2 is super fast now. dolphin-mixtral still takes a long time and uses a ton of CPU -- I wonder if that's because I don't have enough VRAM to run it optimally. But since the errors are gone and I see an improvement, I think I resolved the issue.
GiteaMirror commented 2026-04-12 10:38:19 -05:00
Author
Owner
Copy Link

@PythonGermany commented on GitHub (Jan 26, 2024):

I had the same issue using it like this:

services:
  ollama:
    image: ollama/ollama:latest
    ports:
      - 11434:11434
    volumes:
      - ./data:/root/.ollama
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              capabilities: [ gpu ]

what solved it for me was changing the devices section to this:

          devices:
            - driver: nvidia
              count: 1
              capabilities: [ gpu ]
<!-- gh-comment-id:1912608940 --> @PythonGermany commented on GitHub (Jan 26, 2024): I had the same issue using it like this: ```yaml services: ollama: image: ollama/ollama:latest ports: - 11434:11434 volumes: - ./data:/root/.ollama deploy: resources: reservations: devices: - driver: nvidia capabilities: [ gpu ] ``` what solved it for me was changing the devices section to this: ```yaml devices: - driver: nvidia count: 1 capabilities: [ gpu ] ```
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#936
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?