github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #12217] macOS 15: Ollama defaults to CPU on AMD RX 5700 despite a fully working Metal benchmark #70188

New Issue
Open
opened 2026-05-04 20:37:27 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-05-04 20:37:27 -05:00
Owner
Copy Link

Originally created by @denesfr on GitHub (Sep 8, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12217

What is the issue?

System Details

  • OS: macOS 15.0.1 (Build 24A910)
  • CPU: Intel Core i5-12600KF
  • GPU: AMD Radeon RX 5700
  • RAM: 32 GB
  • System Type: Hackintosh (SMBIOS MacPro7,1)
  • Ollama Version: 0.11.10

Problem Description

Ollama is consistently defaulting to CPU-only inference. The server log explicitly shows library=cpu. It does not appear to even attempt to search for or initialize a Metal-capable GPU, as the "Searching for Metal GPU" log message is absent. This results in 100% CPU usage and 0% GPU usage during inference.

Evidence

1. Ollama Server Log

The server log clearly shows the fallback to the CPU library, even with default environment variables.

time=2025-09-08T14:44:13.644-03:00 level=INFO source=routes.go:1331 msg="server config" env="map[HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:4096 OLLAMA_DEBUG:INFO OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:[http://127.0.0.1:11434](http://127.0.0.1:11434) OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/Users/denesferreira/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NEW_ESTIMATES:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:1 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* [http://127.0.0.1](http://127.0.0.1) [https://127.0.0.1](https://127.0.0.1) [http://127.0.0.1](http://127.0.0.1):* [https://127.0.0.1](https://127.0.0.1):* [http://0.0.0.0](http://0.0.0.0) [https://0.0.0.0](https://0.0.0.0) [http://0.0.0.0](http://0.0.0.0):* [https://0.0.0.0](https://0.0.0.0):* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:false http_proxy: https_proxy: no_proxy:]"
time=2025-09-08T14:44:13.644-03:00 level=INFO source=images.go:477 msg="total blobs: 5"
time=2025-09-08T14:44:13.645-03:00 level=INFO source=images.go:484 msg="total unused blobs removed: 0"
time=2025-09-08T14:44:13.645-03:00 level=INFO source=routes.go:1384 msg="Listening on 127.0.0.1:11434 (version 0.11.10)"
time=2025-09-08T14:44:13.645-03:00 level=INFO source=types.go:131 msg="inference compute" id="" library=cpu variant="" compute="" driver=0.0 name="" total="32.0 GiB" available="18.3 GiB"

2. Proof of Working Metal Compute

Despite Ollama failing to use the GPU, the system's Metal compute capabilities are fully functional, as proven by a successful Geekbench 6 Metal benchmark.

Image

Troubleshooting Steps Taken

  • Verified that the Ollama version is the latest available.
  • Confirmed that the Hackintosh configuration is stable and correctly identifies the GPU for other Metal compute applications (like Geekbench).
  • Tested with both a custom OLLAMA_MODELS environment variable and with the variable unset; the result is the same fallback to CPU.

Relevant log output

OS

macOS

GPU

AMD

CPU

Intel

Ollama version

0.11.10

Originally created by @denesfr on GitHub (Sep 8, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12217 ### What is the issue? ### System Details - **OS:** macOS 15.0.1 (Build 24A910) - **CPU:** Intel Core i5-12600KF - **GPU:** AMD Radeon RX 5700 - **RAM:** 32 GB - **System Type:** Hackintosh (SMBIOS `MacPro7,1`) - **Ollama Version:** 0.11.10 ### Problem Description Ollama is consistently defaulting to CPU-only inference. The server log explicitly shows `library=cpu`. It does not appear to even attempt to search for or initialize a Metal-capable GPU, as the "Searching for Metal GPU" log message is absent. This results in 100% CPU usage and 0% GPU usage during inference. ### Evidence #### 1. Ollama Server Log The server log clearly shows the fallback to the CPU library, even with default environment variables. ``` time=2025-09-08T14:44:13.644-03:00 level=INFO source=routes.go:1331 msg="server config" env="map[HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:4096 OLLAMA_DEBUG:INFO OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:[http://127.0.0.1:11434](http://127.0.0.1:11434) OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/Users/denesferreira/.ollama/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NEW_ESTIMATES:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:1 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* [http://127.0.0.1](http://127.0.0.1) [https://127.0.0.1](https://127.0.0.1) [http://127.0.0.1](http://127.0.0.1):* [https://127.0.0.1](https://127.0.0.1):* [http://0.0.0.0](http://0.0.0.0) [https://0.0.0.0](https://0.0.0.0) [http://0.0.0.0](http://0.0.0.0):* [https://0.0.0.0](https://0.0.0.0):* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:false http_proxy: https_proxy: no_proxy:]" time=2025-09-08T14:44:13.644-03:00 level=INFO source=images.go:477 msg="total blobs: 5" time=2025-09-08T14:44:13.645-03:00 level=INFO source=images.go:484 msg="total unused blobs removed: 0" time=2025-09-08T14:44:13.645-03:00 level=INFO source=routes.go:1384 msg="Listening on 127.0.0.1:11434 (version 0.11.10)" time=2025-09-08T14:44:13.645-03:00 level=INFO source=types.go:131 msg="inference compute" id="" library=cpu variant="" compute="" driver=0.0 name="" total="32.0 GiB" available="18.3 GiB" ``` #### 2. Proof of Working Metal Compute Despite Ollama failing to use the GPU, the system's Metal compute capabilities are fully functional, as proven by a successful Geekbench 6 Metal benchmark. <img width="1201" height="2352" alt="Image" src="https://github.com/user-attachments/assets/bd767bfa-aa0d-4fd5-b565-ac00998fdfd3" /> ### Troubleshooting Steps Taken - Verified that the Ollama version is the latest available. - Confirmed that the Hackintosh configuration is stable and correctly identifies the GPU for other Metal compute applications (like Geekbench). - Tested with both a custom `OLLAMA_MODELS` environment variable and with the variable unset; the result is the same fallback to CPU. ### Relevant log output ```shell ``` ### OS macOS ### GPU AMD ### CPU Intel ### Ollama version 0.11.10
GiteaMirror added the bug label 2026-05-04 20:37:27 -05:00
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#70188
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?