Return to Video

A CUDA Program - Intro to Parallel Programming

  • 0:00 - 0:02
    The typical program looks like this.
  • 0:02 - 0:06
    First, the CPU allocates storage on the GPU.
  • 0:06 - 0:10
    Then, the CPU copies some input data from the CPU to the GPU.
  • 0:10 - 0:16
    Next, the CPU calls some kernels watching these kernels on the GPU that process this data.
  • 0:16 - 0:21
    And finally, the CPU copies the results back to the CPU from the GPU.
  • 0:21 - 0:26
    Now, two of these steps require moving data back and forth between the CPU and the GPU.
  • 0:26 - 0:27
    Is this expensive?
  • 0:27 - 0:33
    Well, in general, you'd like to minimize data transfer between the CPU and the GPU as much as you can.
  • 0:33 - 0:37
    If you're going to move a lot of data and do only a little bit of computation on that data,
  • 0:37 - 0:40
    Cuda or GPU computing probably isn't a great fit for your problem.
  • 0:40 - 0:46
    Generally, we've found that the most successful GPU computing applications do a lot of computation
  • 0:46 - 0:49
    and have a high ratio of computation to communication.
  • 0:49 - 0:51
    They send their data to the GPU.
  • 0:51 - 0:54
    They do a lot of work, and only then, they bring it back.
タイトル:
A CUDA Program - Intro to Parallel Programming
概説:

more » « less
Video Language:
English
Team:
Udacity
プロジェクト:
CS344 - Intro to Parallel Programming
Duration:
0:55
Udacity Robot edited 英語(米国) subtitles for 01-22 A CUDA Program
Udacity Robot edited 英語(米国) subtitles for 01-22 A CUDA Program
Stacy Taylor approved 英語(米国) subtitles for 01-22 A CUDA Program
Stacy Taylor edited 英語(米国) subtitles for 01-22 A CUDA Program
Cogi-Admin added a translation

English subtitles

改訂 Compare revisions