I'm very specific about single versus double precision workloads, and i'd like to know which like.. i don't know, i'm new here, what do you call them? datasets? what i called them in the title. Which ones are reliant on which? Because what i have available, and what i'm looking to help contribute, i want to know whether i need to focus on raw single precision or double precision. right now i have cards very good at double precision but extremely bad at single (Fermi) and i'd like to contribute them where needed at the very least

Very few applications require double precision.

Your best bet is to go to projects that support GPU processing and checkout their forums.
And as for it's performance, it's double precision performance is woeful, single precision is OK (for it's time).

GTX 590
FP32 (float)
1,244.2 GFLOPS

FP64 (double)
155.5 GFLOPS
roughly, one eighth...

Project support for that series may be limited as it's 15 years old, and it is no longer supported by NVidia.

Effective April 2018, Game Ready Driver upgrades, including performance enhancements, new features, and bug fixes, will be available only on Kepler, Maxwell, and Pascal series GPUs. Critical security updates will be available on Fermi series GPUs through January 2019.

---

Grant
Darwin NT.