Is my plan a good idea?

[epixoip]

First, GPU is irrelevant for what you are talking about doing. CPU is faster than GPU for the algorithms you would be able to generate tables for, since it's impossible to gain any acceleration performing straight wordlist attacks against fast hashes.

Second, rainbow tables have NEVER been a viable substitute for straight dictionary attack, only brute force. And even then rainbow tables have been considered deprecated for many years now thanks to GPUs. We stopped using rainbow tables 5+ years ago because they are a waste of disk space, they're absurdly inflexible, they don't scale for shit, and smarter attacks (rule-based, combinator, targeted mask attacks) will win out every time.

Overlooking the idea that arbitrarily-large wordlists are completely braindead, let's say your 100 GB wordlist has about 9 billion words. And let's say you really do have a shitty CPU, and you can only pull 15 MH/s on md5. At 15 MH/s, it would only take you 10 minutes to run through your 100 GB of words. That's nowhere near the 2hrs you estimate. This wordlist can also be re-used for other algorithms and can be used to seed other attacks (rule-based, combinator, hybrid, etc), whereas your massive lookup table can't be used for any of those things.

It's just a terrible idea no matter how you slice it. Crack smarter.