You can also learn more about Metal and MPS on Apple’s Metal page. To get started, just install the latest Preview (Nightly) build on your Apple silicon Mac running macOS 12.3 or later with a native version (arm64) of Python.

In the graphs below, you can see the performance speedup from accelerated GPU training and evaluation compared to the CPU baseline:Īccelerated GPU training and evaluation speedups over CPU-only (times faster) The Unified Memory architecture also reduces data retrieval latency, improving end-to-end performance. This reduces costs associated with cloud-based development or the need for additional local GPUs. This makes Mac a great platform for machine learning, enabling users to train larger networks or batch sizes locally.

Training Benefits on Apple SiliconĮvery Apple silicon Mac has a unified memory architecture, providing the GPU with direct access to the full memory store. The new device maps machine learning computational graphs and primitives on the MPS Graph framework and tuned kernels provided by MPS. MPS optimizes compute performance with kernels that are fine-tuned for the unique characteristics of each Metal GPU family. The MPS backend extends the PyTorch framework, providing scripts and capabilities to set up and run operations on Mac. This unlocks the ability to perform machine learning workflows like prototyping and fine-tuning locally, right on Mac.Īccelerated GPU training is enabled using Apple’s Metal Performance Shaders (MPS) as a backend for PyTorch. Until now, PyTorch training on Mac only leveraged the CPU, but with the upcoming PyTorch v1.12 release, developers and researchers can take advantage of Apple silicon GPUs for significantly faster model training. And on the software side, Apple focuses its energy on its own proprietary Metal graphics API rather than supporting the open Vulkan API or modern versions of OpenGL.In collaboration with the Metal engineering team at Apple, we are excited to announce support for GPU-accelerated PyTorch training on Mac.

This is partly because Windows enjoys a higher usage share than macOS and is thus a bigger target for game developers. Nvidia says that adding native support for Apple Silicon will reduce the app's power consumption and startup time and lead to an 'overall elevated experience,' though it notably doesn't mention gameplay factors like streaming quality or input latency.Īpple Silicon Macs from the M1 to the M1 Ultra all have reasonably powerful GPUs compared to similarly priced and specced PCs, but so far, that hasn't led to an influx of AAA PC titles on the platform. Like most Intel-only Mac apps, GeForce Now could run on Apple Silicon Macs using the Rosetta 2 compatibility layer.

Nvidia released version 2.0.40 of its GeForce Now game-streaming app earlier this week, and among its new features is native support for newer Apple Silicon Macs.

Apple metal gpu support