English subtitles

← Introduction to Dynamic Parallelism - Intro to Parallel Programming

Get Embed Code
3 Languages

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

  1. Hi, my name is Steven Jones, and I'm one of the senior engineers in the CUDA group at Nvidia.
  2. My job is to work on what we think about the CUDA programming model.
  3. I have to figure out what we should and shouldn't do to add to it,
  4. how it should look once we actually do decide to add something,
  5. and of course getting the actual engineering work done under the covers to support the new stuff that we add.
  6. And one of the things I've worked on most recently is what I'm here to tell you about right now.
  7. It's something we call dynamic parallelism.
  8. So, in essence, it's really simple.
  9. Everything you've learned so far about running programs on the GPU
  10. has followed sort of a code processor model where the GPU is fully under control of the CPU.
  11. If you think about it, the CPU can create threads for itself. It cannot work on the GPU like this.
  12. Or it can synchronize to wait for the GPU work to finish.
  13. It sounds pretty simple, but this extra little arrow allows all sorts of new types of problems to be solved on the GPU,
  14. and I'm going to tell you about some of those.