Databases
StackOverflow2013
Postsdbo
POCould someone explain nVidia GPU architecture to me?
 

Note that there are some explanatory texts on larger screens.

plurals
  1. POCould someone explain nVidia GPU architecture to me?
    primarykey
    data
    text
    <p>I am just getting into GPU programming for some of my simulation work, and I've been following this tutorial:</p> <p><a href="http://www.drdobbs.com/parallel/cuda-supercomputing-for-the-masses-part/208401741" rel="nofollow">http://www.drdobbs.com/parallel/cuda-supercomputing-for-the-masses-part/208401741</a></p> <p>However, I'm getting the terminology confused and visiting nVidia's documentation is confusing me further.</p> <p>These terms: streaming multiprocessor, threads, warps, blocks, grid, occupancy, shaders</p> <p>are overwhelming me a little bit.</p> <p>This is my very rough understanding:</p> <p>A GPU is split into multiprocessors (maybe like 2 of them). Each multiprocessor is split into warps, which are split into blocks, which are split into threads. I don't know... I wish I had a diagram. Could someone help me out?</p> <p><strong>EDIT</strong>: Let me explain why I'm asking this. Suppose I'm trying to figure out what a "warp" is. I Google "CUDA warp". The first thing that comes up is <a href="http://developer.download.nvidia.com/CUDA/training/cuda_webinars_WarpsAndOccupancy.pdf" rel="nofollow">http://developer.download.nvidia.com/CUDA/training/cuda_webinars_WarpsAndOccupancy.pdf</a>. So I scroll down and it says:</p> <blockquote> <p>A grid is composed of blocks which are completely independent</p> </blockquote> <p>Okay, I get that.</p> <blockquote> <p>A block is composed of threads which can communicate within their own block</p> </blockquote> <p>Okay...</p> <blockquote> <p>32 threads form a warp</p> </blockquote> <p>Now I'm lost. 32 threads form a warp? Do all threads have to be in the same block? Why have we arbitrarily labeled 32 threads a "warp"? Can I call 12 threads a "satchel"? So if a block can hold 20 threads, does that mean I need 2 blocks to contain a warp? You can see where my confusion is coming from. Everything I look-up depends on knowing other terms, so it's kind of like trying to learn a language by reading the dictionary — it's all circular.</p>
    singulars
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. PTQuestion
    1. USNick
    1. USNick
    plurals
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. PO
      singulars
      1. PTAnswer
    1. VO
      singulars
      1. POCould someone explain nVidia GPU architecture to me?
      1. This table or related slice is empty.
      1. VTUpMod
    2. VO
      singulars
      1. POCould someone explain nVidia GPU architecture to me?
      1. This table or related slice is empty.
      1. VTDownMod
    3. VO
      singulars
      1. POCould someone explain nVidia GPU architecture to me?
      1. This table or related slice is empty.
      1. VTDownMod
    1. COWhat's with the close requests? How can I modify this to prevent it from getting closed?
      singulars
      1. POCould someone explain nVidia GPU architecture to me?
      1. USNick
    2. COI don't know if your question should be closed or not, but why not try reading the CUDA docs (http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html)? It explains everything you're asking. It makes it seem like you haven't done much research prior to asking. Also, the Dr. Dobbs article was a good resource, but it's a bit dated now
      singulars
      1. POCould someone explain nVidia GPU architecture to me?
      1. USalrikai
    3. COThese things get dated so fast... I think I found that from someone else's Stack Overflow question.
      singulars
      1. POCould someone explain nVidia GPU architecture to me?
      1. USNick
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload