RDRAND can do > 500MB/s when invoked by 8 threads running in parallel: http://software.intel.com/en-us/articles/intel-digital-rando... the theoretical maximum is 800MB/s

I wrote an x86-64 asm impl as part of my lightweight Java crypto library (https://github.com/wg/crypto) would be easy to drop into any C program: https://github.com/wg/crypto/blob/master/src/main/asm/rdrand...

Intel released an open source library too, though in tests my asm impl was faster ;-) http://software.intel.com/en-us/tags/20757