GPU Accelerated Training On Mac With Upcoming PyTorch v1.12
Version 1.12 of PyTorch will be bringing the ability of GPU-accelerated model training for Mac users running Apple silicon machines.
Created on May 18|Last edited on May 18
Comment
PyTorch has come out with the announcement that GPU-accelerated training for Mac is coming in the next version of PyTorch. Up until now, Mac users were stuck relying on slow CPU processing or expensive cloud-based solutions. The new update will allow Mac owners running machines fitted with Apple silicon GPUs to get the full power of their hardware for machine learning right on their own device.
By utilizing Metal (Apple’s GPU shader API), PyTorch will be able to send processing instructions straight to an Apple silicon GPU, just like you would with CUDA GPU setups on other systems. Apple silicon GPUs also have direct access to full system memory stores, making Mac a good option for machine learning.
The benefits of GPU accelerated training on Mac
Like with any jump from CPU to GPU processing, this new way to accelerate training on Mac GPUs blows CPU training out of the water.
GPU accelerated training for Mac is available right now if you install the PyTorch nightly build (instructions to install the nightly build are on their site). All you have to do is set your Tensors and Modules to the mps device, just like you normally would with cpu or gpu.
Find out more
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.