Questions from Intel's first academic webcast on many core

Questions from Intel's first academic webcast on many core

Paul Steinberg (Intel)'s picture
  • Is the software manycore challenge just for folks who write operating systems? Or does it affect all programmers?
  • Seems like homogenous manycore will be complicated enough for programmers given the languages and tools availble. Why consider heterogeneous manycore?
  • Where in the curriculum (OS, Architecture, Programming Languages course) and when (Freshman, Sophmore, Senior) do you recommend teaching parallel programming?
  • human brain is the ultimate heterogeneous manycore - is there any bioinspired techniques being considered?
2 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.
Michael Wrinn (Intel)'s picture

Thanks for these interesting questions; let me respond to each of them here:

  • the software manycore challenge will certainly impact all programmers, not just those writing the OS (it will hit them too, of course). Even very basic concepts warrant extra consideration: that for loop is now operates across multiple cores; are the iterations independent? Is the data arranged to be localized? The underlying assumption, a single serial core, has changed.
  • heterogeneous manycore designs are predicted by many (myself included) to become the normal CPU; we see this trend already in announcements of CPUs integrating graphics functions, or communication. While it can seem that the chip guys are playing a nasty practical joke on the software community (and yes, it will be more complicated than the already difficult homogeneous case), this is driven by considerations of power and efficiency -- the physics is inescapable.
  • I'd like to change the question, regarding placement in the curriculum, since I think it presents a false choice. It's not "programming vs parallel programming", it's simply "programming", and it takes place on concurrent hardware. Like the for loop case cited above, the change needs to infuse all aspects of the standard curriculum.
  • On bioinsipired: I happened to catch a talk, this week, on concurrency for neural nets for AI. It seems that the shift in platform design is causing some historically dormant approaches to get another look!

Login to leave a comment.