OpenACC?
Posted: Fri Jul 17, 2015 3:55 pm
OpenACC is for adding some GPU acceleration to non-GPU codes. The most performant solution is to write it for the GPU from the ground up in CUDA or OpenCL, as we’ve done in OpenMM (the key code for FAH GPU cores).
OpenACC is more for legacy codes which could use some GPU acceleration. In that context, it seems very cool, with the claim “Computational Chemistry App: 12x Faster, Under 100 Lines of Code - See more at: http://blogs.nvidia.com/blog/2015/07/13 ... yGfnb.dpuf”. Of course, we’re going for way more than a 12x speedup over a single core.