Skip to content

NVidia’s next gen architecture, Fermi, simplified

NVidia’s been crawling up Intel’s posterior for a couple of years now, moving into the general processing (GP) market, and today, that’s becoming even more of a reality, with the announcement of their Fermi graphics architecture.  It’s a complicated bit, but the main parts are this:

  • An obscene number of stream processors/ cores (512 perhaps).  These are simple cores; they are not analogous to an Intel or AMD CPU core.
  • Groups of cores that:
    • Can be combined to act as a larger, threaded core (in regular computing terms, this is somewhat analogous to where a program is executed)
    • Share 64K of L1 cache (think of it as ram for the individual cores)
  • 768K of L2 cache acts as a buffer for cross talk between the individual L1 cache for each core group.  It also buffers data between higher memory levels (such as the regular RAM for the graphics card).
  • Error checking on RAM (ECC).  This is important, because it reduces or eliminates soft errors caused by a temporary memory issue.

All of this is explained in the video on their site, but it’s a bit more technical, and certainly not watered down at all on the technical details.  Video via NVidia, and some more details and performance info at PC Perspective.

Be Sociable, Share!

Posted in Tech. Tagged with , , .

0 Responses

Stay in touch with the conversation, subscribe to the RSS feed for comments on this post.