The latest version of the AI library 'TensorFlow' optimized for Apple's 'M1' Mac will be released



The latest version of the open source

machine learning (ML) software library TensorFlow was released by Apple on November 18, 2020. This version of TensorFlow officially supports the acclaimed ' M1 ' chip, which has scored high on multiple benchmark results as soon as Apple announced it, allowing you to take full advantage of its performance.

apple / tensorflow_macos: TensorFlow for macOS 11.0+ accelerated using Apple's ML Compute framework.
https://github.com/apple/tensorflow_macos



Leveraging ML Compute for Accelerated Training on Mac --Apple Machine Learning Research

https://machinelearning.apple.com/updates/ml-compute-training-on-mac

Accelerating TensorFlow Performance on Mac — The TensorFlow Blog
https://blog.tensorflow.org/2020/11/accelerating-tensorflow-performance-on-mac.html

Apple releases forked version of TensorFlow optimized for macOS Big Sur | VentureBeat
https://venturebeat.com/2020/11/18/google-releases-new-version-of-tensorflow-optimized-for-macos/

Apple's first proprietary M1 chip has gained tremendous support from various media, showing scores that greatly drain other companies' chips in both computing performance and graphics performance in multiple benchmarks.

Apple's 'M1' -equipped Mac review summary, 'computing revolution' and 'unbelievable feat' storm of acclaim & expectations for Apple Silicon in the future --GIGAZINE



Apple updated its official blog on November 18th, announcing the release of Mac-optimized TensorFlow 2.4. 'For many years, the Mac has been a popular platform for developers, engineers, and researchers. Optimized for Macs with M1 chips, TensorFlow 2.4 takes full advantage of the power of the Mac and significantly improves performance. I will let you. '

According to Apple, TensorFlow 2.4 is optimized for Macs with M1 chips by adopting ML Compute of macOS Big Sur as a framework, and high-speed machine learning (ML) that takes advantage of the performance of M1's 8-core CPU and 8-core GPU. ) Is possible.

The following is training with 5 types of ML algorithms on MacBook Pro + TensorFlow 2.3 (gray) with Intel CPU, MacBook Pro + TensorFlow 2.4 (yellow) with Intel CPU, MacBook Pro + TensorFlow 2.4 (orange) with M1. It is a graph of the time it took to do. In all environments, TensorFlow 2.4 is faster than its predecessor, but the result is the best performance, especially in combination with the M1 chip.



The following is a graph comparing the results of training with TensorFlow 2.3 (gray) and TensorFlow 2.4 (yellow) on the MacPro 2019 model equipped with an Intel CPU. This shows that TensorFlow 2.4 can perform much faster processing than the previous version even in an environment using an Intel CPU.



Regarding this benchmark result, IT news site VentureBeat said, 'It takes 2 seconds to run an old TensorFlow on a 13-inch MacBook Pro with an Intel CPU, but only 1 second with a MacBook Pro with an M1 chip and the latest TensorFlow. In addition, Apple says that even with the MacPro 2019 model equipped with an Intel CPU, it can be done in 2 seconds with optimized TensorFlow instead of 6 seconds with non-optimized TensorFlow. ' I commented.

Google's Panjak Kanwar and Fred Alcober, who are involved in the development of TensorFlow, said, 'This improvement seen in TensorFlow 2 is coupled with Apple's ability to run TensorFlow on iOS with TensorFlow Lite. It shows the breadth of TensorFlow that supports high-speed ML processing on Apple hardware. '

in Software,   Hardware, Posted by log1l_ks