Last week we were happy to work with NVIDIA and PGI this to produce the OpenACC workshop:  Accelerating Applications with OpenACC. Participants learned how to use OpenACC to simplify their GPGPU code.  Using OpenACC directives in standard C and Fortran, programmers insert compiler hints into their code to automatically execute compute-intensive regions of code on accelerators, simplifying code development and improving performance portability.  After the workshop one participant reported his results in an email:

After two days of hacking on a kernel “toy” prototype of the gradient routine from our Discontinuous Galerkin shallow water model, I measured a 10x speedup (30 Gflops vs 3 Gflops) … using OpenACC.

10x speed up… not bad for a two day workshop! Helping to enable these performance improvements were NVIDIA staffers Carl Ponder and Sarah Tariq and PSCer John Urbanic who taught the workshop sessions.  PSC staffers Marcela Madrid,  Roberto Gomez, Anirban Jana and Tom Maiden were also there to assist.

The workshop took place in the David Deerfield Training Center in our office space in Oakland.  We had participants from over 15 different institutions and even a few staff members sat in on the fun.

If you’d like more information about OpenACC or would like to talk to one of our staff members, email us at remarks@psc.edu and we’d love to talk to you about it!