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 #11888] Ollama consistently using CPU instead of Metal GPU on M2 Pro Mac (v0.11.4) #7891

New Issue
Closed
opened 2026-04-12 20:02:39 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-12 20:02:39 -05:00
Owner
Copy Link

Originally created by @boraozkum on GitHub (Aug 13, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11888

What is the issue?

I am experiencing an issue where Ollama (version 0.11.4) consistently uses 100% CPU for model inference on my M2 Pro Mac, despite the system having Metal 3 support. I have tried various troubleshooting steps, but the issue persists.

System Information:

  • Mac Model: M2 Pro
  • macOS Version: 14.6.1
  • Ollama Version: 0.11.4 (installed from official .dmg)

Hardware Information:

  • Processor: Apple M2 Pro
  • Memory: 16 GB
  • Graphics: Apple M2 Pro (Integrated)

Steps Taken & Observations:

  1. Initial Observation: ollama ps consistently reports "100% CPU" for all models (e.g., deepseek-r1:14b).
  2. Debug Logs: Debug logs from Ollama server startup consistently show "library=cpu" even after attempting to force Metal via environment variables:
    • OLLAMA_LLM_LIBRARY=metal (logged in ollama_debug_metal.log)
    • OLLAMA_GPU=metal (logged in ollama_debug_gpu_metal.log)
      The output in both logs remained "library=cpu".
  3. Installation Method: Initially, I had a Homebrew installation, which was uninstalled. The current installation is from the official .dmg file.
  4. Model Re-pull: I removed and re-pulled the deepseek-r1:14b model to rule out corruption. The issue persisted.
  5. Explicit num_gpu in Modelfile: Based on a web search, I created a custom Modelfile for deepseek-r1:14b (named deepseek-r1:14b-gpu) and explicitly set PARAMETER num_gpu 61 (as DeepSeek-R1:14B has 61 hidden layers). After creating and running this new model, ollama ps still reported "100% CPU".

Expected Behavior:
Ollama should utilize the Metal GPU for model inference on my M2 Pro Mac, leading to improved performance.

Actual Behavior:
Ollama consistently uses 100% CPU for model inference, resulting in slower performance.

Additional Context:

  • No Docker installation is present.
  • Ollama executable path: /Applications/Ollama.app/Contents/Resources/ollama

I am happy to provide any further logs or information if needed.

Relevant log output

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.11.4

Originally created by @boraozkum on GitHub (Aug 13, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11888 ### What is the issue? I am experiencing an issue where Ollama (version 0.11.4) consistently uses 100% CPU for model inference on my M2 Pro Mac, despite the system having Metal 3 support. I have tried various troubleshooting steps, but the issue persists. System Information: - Mac Model: M2 Pro - macOS Version: 14.6.1 - Ollama Version: 0.11.4 (installed from official .dmg) Hardware Information: - Processor: Apple M2 Pro - Memory: 16 GB - Graphics: Apple M2 Pro (Integrated) Steps Taken & Observations: 1. Initial Observation: `ollama ps` consistently reports "100% CPU" for all models (e.g., `deepseek-r1:14b`). 2. Debug Logs: Debug logs from Ollama server startup consistently show "library=cpu" even after attempting to force Metal via environment variables: * `OLLAMA_LLM_LIBRARY=metal` (logged in `ollama_debug_metal.log`) * `OLLAMA_GPU=metal` (logged in `ollama_debug_gpu_metal.log`) The output in both logs remained "library=cpu". 3. Installation Method: Initially, I had a Homebrew installation, which was uninstalled. The current installation is from the official `.dmg` file. 4. Model Re-pull: I removed and re-pulled the `deepseek-r1:14b` model to rule out corruption. The issue persisted. 5. Explicit `num_gpu` in Modelfile: Based on a web search, I created a custom Modelfile for `deepseek-r1:14b` (named `deepseek-r1:14b-gpu`) and explicitly set `PARAMETER num_gpu 61` (as DeepSeek-R1:14B has 61 hidden layers). After creating and running this new model, `ollama ps` still reported "100% CPU". Expected Behavior: Ollama should utilize the Metal GPU for model inference on my M2 Pro Mac, leading to improved performance. Actual Behavior: Ollama consistently uses 100% CPU for model inference, resulting in slower performance. Additional Context: - No Docker installation is present. - Ollama executable path: `/Applications/Ollama.app/Contents/Resources/ollama` I am happy to provide any further logs or information if needed. ### Relevant log output ```shell ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.11.4
GiteaMirror added the bug label 2026-04-12 20:02:39 -05:00
GiteaMirror commented 2026-04-12 20:02:40 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 13, 2025):

Server logs will help in debugging.

<!-- gh-comment-id:3184355901 --> @rick-github commented on GitHub (Aug 13, 2025): [Server logs](https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md#how-to-troubleshoot-issues) will help in debugging.
GiteaMirror commented 2026-04-12 20:02:40 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Aug 13, 2025):

Can you paste in the output of ollama -v? I'm wondering if there are some remnants of the old brew installed version floating around.

<!-- gh-comment-id:3185971124 --> @pdevine commented on GitHub (Aug 13, 2025): Can you paste in the output of `ollama -v`? I'm wondering if there are some remnants of the old brew installed version floating around.
GiteaMirror commented 2026-04-12 20:02:40 -05:00
Author
Owner
Copy Link

@anicolao commented on GitHub (Aug 14, 2025):

I've seen this problem on a friend's macbook where it was using rosetta to run the intel binary instead of running a darwin binary. I think we used otool to verify that the ollama binary was the wrong build.

<!-- gh-comment-id:3186301549 --> @anicolao commented on GitHub (Aug 14, 2025): I've seen this problem on a friend's macbook where it was using rosetta to run the intel binary instead of running a darwin binary. I think we used `otool` to verify that the `ollama` binary was the wrong build.
GiteaMirror commented 2026-04-12 20:02:40 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Aug 14, 2025):

cc @dhiltgen

<!-- gh-comment-id:3186398994 --> @pdevine commented on GitHub (Aug 14, 2025): cc @dhiltgen
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#7891
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?