How to change N ,M,K in DGEMM in GPU linpack?

Accelerated Computing CUDA CUDA Programming and Performance
1

I test GPU linpack with 384 nodes GPU Kepler for our customer .
How to change N ,M,K in DGEMM in GPU linpack?
What size of N,M,K is good performance for linpack?
Where can find tuning performance document for gpu linpack ?
Thank you.

2

Please ask your customer to contact NVIDIA to get the Linpack version optimized for Kepler ( it is different from the public available one).

3

I have the FERMI_HPL_V.15, but I´m working with 2 Keplers GPU (GTX680), do you have any idea if it should work as well with that Linpack?

Sorry if it is a dumb question, I new in this kind of things :s. Thanks