Does Your M1 Mac Actually Accelerate TensorFlow?

What it does — This repo provides a trio of Jupyter notebooks that train standard machine learning models—two deep learning benchmarks using TensorFlow (TinyVGG on CIFAR-10 and EfficientNetB0 on Food-101) and one classical ML test with scikit-learn’s RandomForestClassifier—specifically to measure performance on Apple Silicon chips. It functions largely as a setup validator for getting TensorFlow running on M1, M1 Pro, M1 Max, M1 Ultra, and M2 Macs using Apple’s Metal GPU framework. Results are collected in a dedicated results directory for comparison against other hardware.

The interesting bit — The project is less about novel architectures and more about acting as a practical smoke test for Apple’s entire ML software stack, including the tensorflow-macos fork and the tensorflow-metal plugin. For developers wondering whether their new Mac’s GPU cores are genuinely being leveraged by TensorFlow, this offers a reproducible way to verify.

Key highlights —

• Three benchmark notebooks covering computer vision and tabular data
• Targets Apple Silicon specifically (M1 through M2)
• Leverages tensorflow-metal for GPU acceleration on Mac
• Includes a results directory for performance comparisons
• README prioritizes environment setup validation

Caveats —

• The README is overwhelmingly focused on installation steps rather than architectural insights
• Benchmarks rely on fairly conventional model architectures (TinyVGG, off-the-shelf EfficientNetB0)
• Python 3.8 is noted as the “most stable” version, which may feel dated

Verdict — Worth a look if you need to prove your Apple Silicon Mac can run TensorFlow workloads or want comparable benchmark numbers across devices. Skip it if you’re hunting for novel model architectures or non-Mac optimization tricks.