Google Starts Selling Edge TPUs

Discussion in 'HardForum Tech News' started by AlphaAtlas, Mar 5, 2019.

  1. AlphaAtlas

    AlphaAtlas [H]ard|Gawd Staff Member

    Messages:
    1,713
    Joined:
    Mar 3, 2018
    Back in 2016, Google revealed their custom built Tensor Processing Unit chips that were explicitly designed for machine learning tasks, and just last year, they started renting out cloud-based access to updated versions of that AI hardware. These monster ASICs are squarely aimed at ML training tasks, but during that same period, Google also released a small, low power Edge TPU designed to run algorithms the big chips train a little closer to home. Previously, these "Edge" ML chips were only available to rent through Google, but, an Alphabet spin-off called Coral just started selling the Edge TPUs through Mouser. Interested parties can buy self-contained development boards complete with an ARM CPU, an integrated GPU, I/O, and Ethernet for $149, while a USB accessory akin to an Intel Compute Stick will set you back $75. A PCI-E based accelerator and a 40mm x 40mm "System on Module" are said to be coming sometime in 2019, but there's no word on when, or if, Google will ever sell the bigger TPU ASICs as discrete co-processors.

    AI is pervasive today, from consumer to enterprise applications. With the explosive growth of connected devices, combined with a demand for privacy/confidentiality, low latency and bandwidth constraints, AI models trained in the cloud increasingly need to be run at the edge. Edge TPU is Google's purpose-built ASIC designed to run AI at the edge. It delivers high performance in a small physical and power footprint, enabling the deployment of high-accuracy AI at the edge.
     
    DejaWiz likes this.
  2. bobdabilder

    bobdabilder Limp Gawd

    Messages:
    292
    Joined:
    Oct 7, 2009
    That has the potential to be pretty cool.
     
    steakman1971 likes this.
  3. Cactusj

    Cactusj n00b

    Messages:
    46
    Joined:
    Jun 4, 2018
    How long until someone has RetroPi running on this?
     
  4. DejaWiz

    DejaWiz Oracle of Unfortunate Truths

    Messages:
    18,641
    Joined:
    Apr 15, 2005
    Not fast enough
     
  5. TheHobbyist

    TheHobbyist Hugs Hard Johnnies [H]ard

    Messages:
    456
    Joined:
    Apr 8, 2008
    But can it run Crysis?
     
    GoldenTiger and tempertantrum like this.
  6. lollerwaffle

    lollerwaffle Gawd

    Messages:
    666
    Joined:
    Feb 3, 2008
    I'm wishing for AI to make a real difference in games. Can it make for smarter, more believable, more diverse enemies in SP games? I'll always have a PCI-E slot ready for that sort of thing.
     
    Domingow likes this.
  7. SamuelL421

    SamuelL421 Limp Gawd

    Messages:
    140
    Joined:
    Jun 3, 2016
    Someday maybe, but AI as it currently exists can only create unbelievable super bots. AFAIK, you can train an AI that will exploit everything short of outright cheating and it can be godlike for a particular set of tasks in a game. The AI bot could analyze many outcomes of an action to predict the best move it should make (like the Starcraft or Go AI's) - but it won't be random, creative, or show any sort of real problem-solving ability.

    TLDR: AI can currently make a bot that will kick your ass, but it won't fool you into thinking its a real person.
     
  8. clockdogg

    clockdogg Gawd

    Messages:
    849
    Joined:
    Dec 12, 2007
    So....AI gamebots are the ultimate 'factory-sponsored' cheat codes.
     
    Marees likes this.