Got a YouTube account?

New: enable viewer-created translations and captions on your YouTube channel!

English subtitles

← Theoretical Peak Bandwidth - Intro to Parallel Programming

Get Embed Code
2 Languages

Showing Revision 4 created 05/25/2016 by Udacity Robot.

  1. So 2,508 megahertz is 2,508 times 10 to the 6 clocks per second.
  2. Memory bus of 128 bits is equal to 16 bytes per clock.
  3. So if you multiply this out,
  4. the maximum theoretical peak bandwidth of the memory system
  5. is going to be just over 40 gigabytes per second.
  6. And as a rough rule of thumb,
  7. anytime we achieve something like 40 to 60% of the memory bandwidth,
  8. we'd say, "Well, that's doing okay. It's not great. There's probably room for improvement."
  9. If you can get into the 60 to 75% range, that's doing pretty well.
  10. You might not be able to improve on that.
  11. And any time you get over 75%, we would consider that excellent, okay?
  12. You'll never achieve this theoretical peak bandwidth on any real substantial code, okay?
  13. This is literally just what you get from multiplying out the clock rate and the memory bus,
  14. and a real code is going to have an additional overhead.
  15. So if you can get over 75% of that, you're doing really well,
  16. and you probably don't need to optimize the memory further.
  17. So how well are we doing in this code?