kavisek / new-mac-performance-testing

This is a repo on performance testing a back for python and machine learning development.

New Mac Performance Testing

This is a epo for performance testing a backend for Python and machine learning development.. Contains some simple Python scripts.

• Python Matrix Multiplication
• GPU Benchmarking (Pytorch) (Notebook)

# This example uses python 3.11
cd app 
poetry install
poetry shell
python ./scripts/matrix_multiplication.py

2023 M3 Max MacBook Pro

Results: Matrix multiplication of size 16000x16000 took 16.08 seconds

View Core Utilization

1. Open Activity Monitor
2. Right Click the Activity Monitor Dock Icon
3. Select Monitor -> Show CPU History

{ width=25% }

View GPU Utilization

1. Open Activity Monitor
2. Select Window in the Menu Bar -> GPU History

Machine Learning Performance (Pytorch Matrix Multiplication)

You can view the notebook here. This notebook runs 30 samples of matrix multiplication of size 50,000x50,000 tensor on both the CPU and GPU. The GPU is ~10x faster than the CPU for machine learning workloads using Pytorch.

I am currently using a 2023 MacBook Pro with a M1 Max Ultra Configuration with 16 core CPU and 40 core GPU and 128GB of unified memory (RAM).

cd app
poetry install
poetry run jupyter lab 

Here are the notebook results:

References

• https://github.com/alexziskind1/machine_tests/tree/main/python/matrix_multiplication
• https://github.com/PlummersSoftwareLLC/Primes
• https://benchmarksgame-team.pages.debian.net/benchmarksgame/program/mandelbrot-python3-7.html