home - all - feeds - topics | technology - microblogs - databreaches - movies - Beachboys - diy - camping - colombia - solarpunk - tvmovies - linux - microblogs - onehundredninetysix - pravda_news - canada - historymemes - edmonton - pathofexile - fauxmoi - languagelearning - more »
log in or sign up | preferences |
quokk.au site icon
fcalva@cyberplace.social
  • overview
  • comments
  • submitted
  • fcalva

    fcalva@cyberplace.social

    Lisp enjoyer, occasional code golfer.I self host what i can, because i can.As for the rest, i invite you to read my timeline.

    joined 2 years ago
    sorted by: new top controversial old
    China solves 'century-old problem' with new analog chip that is 1,000 times faster than high-end Nvidia GPUs in c/technology@lemmy.world
    [–] fcalva@cyberplace.social 12 points 1 year ago (2 children)

    @carrylex @kalkulat If you zoom in, you will see that it is in fact a real photo. It's just someone who thought putting a CPU upside down was a good idea.

  • permalink
  • fedilink
  • source
  • parent
  • context
  • China solves 'century-old problem' with new analog chip that is 1,000 times faster than high-end Nvidia GPUs in c/technology@lemmy.world
    [–] fcalva@cyberplace.social 2 points 1 year ago

    @Treczoks @flemtone Thing is, the final LLM inference is usually done at reduced precision. 8-16 bits usually, but even 4bits or lower with different layers of varying precision.

  • permalink
  • fedilink
  • source
  • parent
  • context
  • China solves 'century-old problem' with new analog chip that is 1,000 times faster than high-end Nvidia GPUs in c/technology@lemmy.world
    [–] fcalva@cyberplace.social 1 point 1 year ago

    @TheBlackLounge @kalkulat LLM inference is definitely theoretically possible on analog chips. They just may not scale :v

  • permalink
  • fedilink
  • source
  • parent
  • context
  • piefed
    0.5.3
    mlmym
    dotzip.1.9
    quokk.au
    modlog