Homebrew version of ollama can't detect GPUs.

[pcastellazzi]

brew config AND brew doctor output OR brew gist-logs <formula> link

brew config
HOMEBREW_VERSION: 5.0.8
ORIGIN: https://github.com/Homebrew/brew
HEAD: b473452cf25e5f70522229d1472948584b74736b
Last commit: 4 days ago
Branch: stable
Core tap JSON: 01 Jan 04:26 UTC
HOMEBREW_PREFIX: /home/linuxbrew/.linuxbrew
HOMEBREW_CASK_OPTS: []
HOMEBREW_DISPLAY: :0
HOMEBREW_DOWNLOAD_CONCURRENCY: 32
HOMEBREW_EDITOR: vim
HOMEBREW_FORBID_PACKAGES_FROM_PATHS: set
HOMEBREW_MAKE_JOBS: 16
HOMEBREW_NO_ANALYTICS: set
HOMEBREW_NO_AUTO_UPDATE: set
HOMEBREW_NO_ENV_HINTS: set
Homebrew Ruby: 3.4.8 => /home/linuxbrew/.linuxbrew/Homebrew/Library/Homebrew/vendor/portable-ruby/3.4.8/bin/ruby
CPU: 16-core 64-bit rocketlake
Clang: N/A
Git: 2.43.0 => /bin/git
Curl: 8.5.0 => /bin/curl
Kernel: Linux 6.6.87.2-microsoft-standard-WSL2 x86_64 GNU/Linux
OS: Ubuntu 24.04.3 LTS (noble)
WSL: 2 (Microsoft Store)
Host glibc: 2.39
Host libstdc++: 6.0.33
/usr/bin/gcc: 13.3.0
/usr/bin/ruby: N/A
glibc: N/A
gcc@12: N/A
gcc: N/A
xorg: N/A



brew doctor
Your system is ready to brew.



brew gist-logs ollama
Error: No logs.

Verification

• My brew doctor output says Your system is ready to brew. and am still able to reproduce my issue.
• I ran brew update and am still able to reproduce my issue.
• I have resolved all warnings from brew doctor and that did not fix my problem.
• I searched for recent similar issues at https://github.com/Homebrew/homebrew-core/issues?q=is%3Aissue and found no duplicates.
• My issue is not about a failure to build a formula from source.

What were you trying to do (and why)?

I want to use ollama to run a local LLM model.

What happened (include all command output)?

/home/linuxbrew/.linuxbrew/bin/ollama serve 2>&1 | grep "inference compute"
time=2026-01-02T06:12:04.010-03:00 level=INFO source=types.go:60 msg="inference compute" id=cpu library=cpu compute="" name=cpu description=cpu libdirs=ollama driver="" pci_id="" type="" total="15.5 GiB" available="14.9 GiB"

GPU is not detected.

What did you expect to happen?

/usr/local/bin/ollama serve 2>&1 | grep "inference compute"
time=2026-01-02T06:12:49.162-03:00 level=INFO source=types.go:42 msg="inference compute" id=GPU-49b4f306-b610-0f3c-ff52-e2c0a86020bd filter_id="" library=CUDA compute=7.5 name=CUDA0 description="NVIDIA GeForce RTX 2080 Ti" libdirs=ollama,cuda_v13 driver=13.0 pci_id=0000:01:00.0 type=discrete total="11.0 GiB" available="9.8 GiB"

Using ollama from https://ollama.com, GPU is correctly detected.

Step-by-step reproduction instructions (by running brew commands)

# install
brew install ollama
# use the version installed by Homebrew
/home/linuxbrew/.linuxbrew/bin/ollama serve 2>&1 | grep "inference compute"

# install
curl -fsSL https://ollama.com/install.sh | sh
# use the version installed by the script
/usr/local/bin/ollama serve 2>&1 | grep "inference compute"

From what i can gather the version of ollama provided by homebrew does not include any GGML backends.

[Pnkcaht]

Response

After reviewing the formula, Homebrew builds Ollama using a plain Go build without enabling any GGML GPU backends (such as CUDA, ROCm, or Metal).
This explains why the Homebrew-installed binary only detects CPU devices, while the official installer provides GPU-enabled builds.
The behavior seems expected, but it is currently undocumented. I’ve opened a PR to add caveats to the formula clarifying that GPU acceleration is not included and pointing users to the official installer if GPU support is required.

Code

def caveats
  <<~EOS
    This formula builds Ollama without GPU acceleration
    (e.g. CUDA, ROCm, or Metal).

    If you require GPU support, please use the official installer:
      https://ollama.com/download
  EOS
end

[SMillerDev]

Homebrew uses go build to build ollama. It seems this does not automatically enable GPU acceleration. It might be good to look into building the formula with CMake instead, a pull request for that would be appreciated.

[Pnkcaht]

Homebrew uses go build to build ollama. It seems this does not automatically enable GPU acceleration. It might be good to look into building the formula with CMake instead, a pull request for that would be appreciated.

I had already sent a PR regarding what I mentioned, but I will review what you said and send another PR.

[cho-m]

NVIDIA CUDA is proprietary so we cannot build support in ollama formula due to our FOSS requirement - https://docs.brew.sh/License-Guidelines

Only Vulkan support could be enabled, but that may only work with AMD/Intel GPUs - https://github.com/ollama/ollama/blob/main/docs/development.md#linux

On macOS, we do allow using system libraries like Metal. According to upstream, Metal is enabled by default on Apple Silicon and needs to build bundled GGML for Intel macOS https://github.com/ollama/ollama/blob/main/docs/development.md#macos-apple-silicon

---

Though, given original issue is on NVIDIA GPU, the above just means the behavior is expected. Maybe a Cask for the NVIDIA variant could be provided, but would depend on state of Linux Cask support