Main point : Tesla K20 has some unique technologies that make it easier than ever before

  * hyper-Q : the key benefit is that there are a lot of MPI codes out there that had a hard time porting to GPUs because the way the way it’s written
   - They typically write MPI codes with small job sizes for CPU cores, but GPUs need large job sizes.
   - Hyper-q allows up to 32 Mpi jobs to run on a single GPU

  * Dynamic parallelism: the key benefit is gpu codes look a lot like cpu codes.. The code size shrinks and looks more natural.