Feb 16, 2021 20:00:00

As a result of verifying whether M1 equipped Mac is suitable for deep learning

Apple's proprietary

Apple Silicon for Mac, the first memorable chip, the M1 , has received acclaimed reviews from foreign media, including the 'computing revolution' and 'incredible feats.' .. Better Data Science , a data science-related media, has verified that the M1 chip is 'suitable for deep learning?'

Are The New M1 Macbooks Any Good for Deep Learning? Let's Find Out | Better Data Science
https://www.betterdatascience.com/m1-deep-learning/

The M1 chip is a chipset that combines an 8-core CPU, 8-core GPU, 16-core neural engine, etc., and it has become clear that the CPU performance and GPU performance alone are higher than the previous generation Mac. I will.

The following shows the results of benchmark testing of the CPU and GPU of the M1-equipped MacBook Pro (bottom) that appeared in 2020 and the Intel processor-equipped MacBook Pro (top) that appeared in 2019 using Geekbench 5 . The single-core, multi-core, and OpenCL scores are all overwhelming for the M1-equipped MacBook Pro.

Multiple benchmarks have shown that the M1 chip is high-performance, but Better Data Science verifies how fast the M1 chip works for deep learning-related tasks. I am.

However, the M1 chip isn't compatible with all data science libraries, 'for example, getting TensorFlow version 2.4 to work properly on an M1-powered Mac isn't as easy as it sounds,' Better Data Science explains. doing.

Therefore, Better Data Science notes that the benchmark results shown below only show the 'average training time' when deep learning is performed on each Mac, and are 'not scientific'. I am writing.

The following is the time taken to train a

neural network classifier with epoch count '10' using the MNIST database, which is treated as a rudimentary dataset in deep learning, with Google Colab, which Google provides for free. Compared graph. Better Data Science wrote, 'This result is a bit disappointing,' as the result is that the M1-powered MacBook Pro takes longer to train than the Google Colab.

The following shows the time taken to train the same neural network as above using

Fashion-MNIST with clothing images added to the MINST database. The result is the same as when using the MINST database, but so far it is just a test result when using a simple data set.

How long does it take to train a neural network classifier with three convolutional layers using the CIFAR-10 , a 'more advanced dataset' containing various types of photographs such as animals and vehicles? The graph below is summarized. As a result of using more complex neural networks and datasets than those used above, training times have also changed significantly, and training on an M1-powered MacBook Pro is faster than training in a Google Colab CPU environment. .. However, training takes less time in Google Colab's GPU environment than in a MacBook Pro with an M1.

Better Data Science says, 'The M1 chip offers better performance, no overheating, and better battery life. Still, it's hard to recommend to anyone interested in deep learning.' Is about twice as performant as an Intel-based Mac, but it's not a machine built for deep learning. Don't get me wrong. The MacBook Pro can be used for basic deep learning tasks. You can, but if you do deep learning on a daily basis, there are better machines in the same price range, 'he said, saying that there are better machines for deep learning.