Von einem Google-Ingenieur gegründetes chinesisches Startup behauptet, einen eigenen TPU-Chip für KI entwickelt zu haben – ein benutzerdefinierter ASIC, der Berichten zufolge 1,5-mal schneller als Nvidias A100-GPU aus dem Jahr 2020 und 42 % effizienter ist

    https://www.tomshardware.com/tech-industry/chinese-startup-founded-by-google-engineer-claims-to-have-developed-its-own-tpu-reportedly-1-5-times-faster-than-nvidias-a100-gpu-from-2020-42-percent-more-efficient

    Share.

    5 Kommentare

    1. PhantomGaming27249 on

      So it’s a bit worse than a h100. Totally usable but not earth shaking but definitely a good GPU.

    2. Secret_Wishbone_2009 on

      It is ok to come with silicon that will become more competitive but what about the ecosystem around it?

    3. If your TPU is only 1.5× faster than a GPU you’re doing something wrong. It’s supposed to be many times faster than that

    4. >“reducing power consumption to 75 per cent using a manufacturing process that is an order of magnitude lower than that of leading overseas GPU chips.“

      Sounds like total fiction, that would be way into breaking the laws of physics territory.

    Leave A Reply