Which GPU is good in my scenario ?

My application is memory bound, and involves lots of modular arithmetic, including inverse modulo operations.
Can I get some suggestions for choosing best Graphics card suiting my requirement ?

What all should I look for apart from memory bandwidth?

The GPUs with the highest memory bandwidth are currently found in the Geforce 700 series, namely the GTX 780 Ti and the GTX Titan Black: http://en.wikipedia.org/wiki/GeForce_700_series