In the modern era of high-performance computing, the term “Performance Portability” elicits a wide range of reactions… and a wide range of definitions. At their heart, the terms Performance and Portability are somewhat adversarial— or at the very least exist on independent axes. For the purposes of this panel, we’ll define Performance Portability as “the ability to run a mostly-single-source code mostly-efficiently most of the time on most computer architectures” and attempt to better understand expert perspectives on the topic.

Expect questions and discussion on:

  • The many types of “performance” and how much each type matters to different communities (code developers, code users, sysadmins, etc). Is there one right answer?
  • The current state of the art in portability and if/how panelists see this changing in the near future.
  • The impact that modern performance portability tools (e.g. Kokkos) have on old/legacy codes.
  • The pros/cons of outsourcing portability concerns to an external dependency - code readability, version compatibility, additional indirection, etc.

Panelists

  • Carl Pearson (Sandia)
  • Joel Rhodes III (Studsvik Scandpower)
  • Joe Zerr (LANL)
  • Braxton Cuneo (Seattle Univ.)

Discussion

To join the conversation, you must be logged in and registered for the meeting.

Register NowLog In