Apple’s M-series chips, particularly the M4, have significant untapped potential for scientific computing, especially in protein folding. While these chips are powerful and feature a unified memory architecture, most academic software heavily relies on CUDA, limiting their use in fields like bioinformatics. The MLX framework, designed specifically for M chips, optimally utilizes this unified memory, which can significantly enhance performance for memory-intensive tasks. Traditional frameworks like PyTorch and TensorFlow struggle with memory bottlenecks due to their reliance on separate CPU and GPU memory pools. The article highlights the author’s experience porting OpenFold3, an open-source version of AlphaFold3, to MLX. The transition required rewriting CUDA-specific operations, such as triangle attention, into MLX-compatible code. The performance results on an M4 MacBook Air were promising: small proteins took 20-30 seconds, medium proteins around 1:30, and large proteins about 3 minutes for inference. This contrasts sharply with longer wait times on CPU-only setups. The author emphasizes that anyone using MLX version 0.5.0 or later can expect improved functionality. The shift in computational biology is now more about software efficiency than access to expensive hardware. With MLX, researchers can leverage existing Apple devices to achieve competitive performance in protein folding tasks. This opens up opportunities for broader participation in scientific computing without the traditional barriers of high-cost GPU setups.