Speed up Stylegan3 training?


Yes, I am a noob and hack of a developer. My background is mostly pillaging scripts made by competent creators to help me automate or schedule my needs for various legacy systems at companies.

I hope I got the right forum.

I have been playing around Stylegan3…and finally got it to train (train.py). (Found on github)
It is going to take 73.6 days to complete. That is a 4090 running for ~1766 hours.
In 73 days, I have a strange feeling much will change anyhow.

Since there is no way I am leaving my computer on for that long (plus need it for other work).

I am thinking of setting checkpoints up, so I can run it, and let it resume later. This will likely take over a year
Perhaps, reduce the size of the training images (512x512)

Q1: Any thoughts of something else I can do to make it faster
Q2: It uses PyTorch. If supported, would TensorFlow make any difference?
Q3: Has anyone already done this and would they please share their experiences (positive or negative)

13th Gen Intel(R) Core™ i9-13900KF
NVIDIA GeForce RTX 4090
64GB Memory
2TB Storage / 4TB SSD / 2TB HDD

If I can provide more info - please let me know.
Thank you!