1. Forum
  2. USENET
  3. comp.arch

  • AI Compute Extensions by the x86 Ecosystem Advisory Group

    From Thomas Koenig@tkoenig@netcologne.de to comp.arch on Fri Jun 19 16:41:36 2026
    From Newsgroup: comp.arch

    Intel and AMD call it ACE. Details to be found at https://x86ecosystem.org/resource/ai-compute-extensions-ace-specification/

    One interesting data type I had not come across before is a fixed
    point type with implicit exponent scaling a signed 8-bit number
    * 2**(-6). With this, you cannot have a normal accumulation
    of the same data type; they use 32-bit float instead.
    --
    This USENET posting was made without artificial intelligence,
    artificial impertinence, artificial arrogance, artificial stupidity,
    artificial flavorings or artificial colorants.
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From MitchAlsup@user5857@newsgrouper.org.invalid to comp.arch on Fri Jun 19 22:07:13 2026
    From Newsgroup: comp.arch


    Thomas Koenig <tkoenig@netcologne.de> posted:

    Intel and AMD call it ACE.

    And I thought "ACE is the place with the helpful Hardware Man" ...

    Details to be found at https://x86ecosystem.org/resource/ai-compute-extensions-ace-specification/

    One interesting data type I had not come across before is a fixed
    point type with implicit exponent scaling a signed 8-bit number
    * 2**(-6). With this, you cannot have a normal accumulation
    of the same data type; they use 32-bit float instead.

    A lot of AI is trying to figure out how to get the job done without
    moving so much data around. I suspect, long term, that this will be
    shown to be a poor choice ...
    --- Synchronet 3.22a-Linux NewsLink 1.2