K80 single precision


I apologize for such a basic question, but I’m new to GPU accelerated computing.

In the literature for K80 cards, the following information is presented:

Up to 2.91 Teraflops double-precision performance with NVIDIA GPU Boost
Up to 8.73 Teraflops single-precision performance with NVIDIA GPU Boost

How do I select which precision performance for a given job? Or is it just the software analysis routines that I’m using that will determine this?

I am asking because I am running code that is single precision and I wanted to make sure it was running as fast as possible on AWS p2 instances with K80s.

Along those lines - are there any tweaks that I can make to improve the performance for single precision calculations on K80s?

Thank you,

single precision is operations on C float type, double precision - on C double type. if you are running existing software, you can’t change its precision without modification of sources