Apple is adding support for NVIDIA CUDA to its machine learning framework MLX, which is designed for Apple Silicon chips. This breakthrough will provide AI developers with unprecedented flexibility and cost advantages.
According to Appleinsider, developers can now use the MLX framework to develop AI applications on Macs equipped with Apple Silicon and export their code to run on NVIDIA GPUs or server environments that support CUDA. This capability means developers can build model prototypes on macOS and seamlessly migrate to the NVIDIA platform during deployment, fully utilizing its powerful computing capabilities.
Previously, MLX was heavily dependent on Apple's own Metal framework, limiting its operation to the macOS system. Developers who wanted to deploy in a broader environment had to purchase expensive NVIDIA hardware for adaptation and testing, increasing development costs and barriers.
This CUDA support was led by GitHub developer @zcbenz, who spent several months developing, splitting, and integrating related modules, finally merging the code into the main branch of MLX. It is worth noting that this project does not mean native CUDA support on Macs, nor can it allow MLX on Macs to directly access NVIDIA GPUs through external graphics cards. Its core value lies in "code export compatibility," paving the way for cross-platform deployment.
For developers, this update offers the most direct benefit in terms of cost control: they can complete the development process on performance-rich but more affordable Apple Silicon Macs, and only transfer to expensive NVIDIA hardware when necessary for deployment or training large models. For startup teams and individual developers, this is undoubtedly a significant reduction in entry barriers.
Additionally, due to the powerful computing power of NVIDIA hardware in AI training tasks, MLX is expected to achieve significantly better performance after migrating to the CUDA platform, thus greatly improving training efficiency and model accuracy.
This compatibility expansion retains the efficient experience of Apple Silicon development while expanding the openness of the deployment level, and may become an important turning point for the MLX framework to enter a broader application ecosystem.