Page 1 of 1

Testing with fp32 GPU?

Posted: Mon Oct 23, 2017 6:25 am
by Klausb

to debug my application, I am wondering whether there's a solver which does the fp64 (double) computation on the CPU?

An alternative (possibly better) approach would be to convert the matrix, rhs and x to float, do the computation in fp32 on the GPU and convert it back to double (fp64). Is that possible?

My starting point is always double (fp64).