
There are tons of good expositions on the FFT-way, e.g. check out
"Prime Numbers - A Computational Perspective" by Crandall-Pomerance see ch. 9, Fast algorithms for large-integer arithmetic
In that timeframe, the main technical advances that helped GIMPS get more efficient were being made by Richard E. Crandall (https://en.wikipedia.org/wiki/Richard_Crandall), who also wrote Prime Numbers: A Computational Perspective with Carl Pomerance (https://en.wikipedia.org/wiki/Carl_Pomerance). If these folks are trying to help governments keep a monopoly on expertise about number theory and the difficulty of the RSA problem, they seem to be doing a pretty bad job of it.
also the laddering scheme used for the 'exponentiation' can be in different forms, the left-to-right or right-to-left form or even a Montgomery ladder...
i seemed to recall there is a 'fastest known' method for generating fibonacci or lucas numbers... but google is not helping me.
pretty sure i had seen it in this book: https://www.amazon.co.uk/Prime-Numbers-Computational-Carl-Po... i'll have to check when i have access to it next. :)