Local training with M1 vs AMD 6900xt performance

FYI I’ve been doing this course by running all training locally. I am really impressed by the M1 Pro performance vs my Desktop. The Tensorflow-macos using metal api is finally stable enough to do all the training for this course series so far. This is some training statistics for training time in W3 homework:

M1 CPU 10-core (macOS 12.4): 101.4 seconds
M1 GPU 16-core (Metal api macOS 12.4): 40.8 seconds

I9 12900k CPU (macOS 12.4): 212.3 seconds
AMD 6900xt GPU(Metal api macOS 12.4): 90 seconds
AMD 6900xt GPU (ROCM 5.2 Ubuntu 20.04): 51.4 seconds

Just wanted to post in case there are other students out there with m1 Macs you can probably get better performance running locally vs the free Jupyter cloud!

Thanks for sharing your benchmarks