higher precision arithmetic for NVIDIA GPUs

[Stefan Lang]

Dear all,

current nvidia graphics card allow floating point arithmetic
with float precision, there is no double available.
For numerical algorithms I want to run on these cards, this
lack in precision leads to unreliable results.
Now my questions:
1. Can libgmp easliy be adapted completely or some part (I only need
some BLAS like routines)  for these cards?
2. Does anyone know an efficient way to implement an arithmetic with higher 
    precision (using the float type already there) by myself?

Best,
Stefan Lang