A Milliwatt Of DOOM

The seminal 1993 first-person shooter from id Software, DOOM, has become well-known as a test of small computer platforms. We’ve seen it on embedded systems far and wide, but we doubt we’ve ever seen it consume as little power as it does on a specialized neural network processor. The chip in question is a Syntiant NDP200, and it’s designed to be the always-on component listening for the wake word or other trigger in an AI-enabled IoT device.

DOOM running on as little as a milliwatt of power makes for an impressive PR stunt at a trade show, but perhaps more interesting is that the chip isn’t simply running the game, it’s also playing it. As a neural network processor it contains the required smarts to learn how to play the game, and in the simple circular level it’s soon picking off the targets with ease.

We’ve not seen any projects using these chips as yet, which is hardly surprising given their niche marketplace. It is however worth noting that there is a development board for the lower-range sibling chip NDP101, which sells for around $35 USD. Super-low-power AI is within reach.

36 thoughts on “A Milliwatt Of DOOM

  1. With all this talk of AI lately, and now low power AI, I’m reminded of the quote “I’m more afraid of a computer failing the Turing test on purpose than passing it.”

    On a more related note, the average human brain runs on about 20 watts. Seems like there’s enough available to run a few augmentation chipsets if they’re only drawing a few milliwatts.

    1. We are an at rest 100 ish watt biological machine with neurons / biological AI scattered all around the body. The brain is where the We and I concept exists along with the sensory processing centres but the whole body is responsible for basically its own small part of the holy show that is a human.
      Every muscle is controlled by neurons within the muscle. glands and organ are the same.
      Don’t think of AI chips as brains, certainly not the concept of I and consciousness yet. Think of it more like the basic input filtration akin to a neuron in a muscle where its just idle until the right neurotransmitter / data comes along and simulates an action. An actual genuine thinking machine isn’t impossible its just not that probable anytime soon. Innovation isn’t part of what modern AI can do, it can only do pattern recognition based on data sets, it has no ego or motivation via consciousness to really truly explore and form new concepts and actively direct its efforts to acquiring new information.
      We will get there. We just won’t get there tomorrow or next Tuesday.

  2. Reading comprehension test… :)

    The article does nowhere say that the NDP200 “runs” doom, but instead it “plays” doom. It is a neural network accelerator after all, and was trained to “play” an implementation of doom. (A narrowly defined subset of it, it seems).

    Here is the original tweet the article is based on:

    https://twitter.com/dylan522p/status/1628110148278562816

    The Doom implementation in question seems to “run” on a Raspberry PI and would certainly consume more than 1mW of power.

      1. Well, maybe. Consider a military drone. I think it would be quite the challenge to make it work. A lot of training data to start with. Unless you have millions of drones laying around. And where do you get that training data? Drone pilots.

        Now, you need to take the infant AI by the hand, prevent it from overfitting or underfitting, correct when required.

        Now the tricky part, after learning from human pilots, it enters the chaos of battle. There are so many parameters, you simply can’t always evade everything. So the AI might fly perfect, make the best choices, and still gets shot down. Machine learning is just not the right tool for that job. Too noisy data.

    1. Interestingly enough, memory is one of the areas of study where we are making progress in neural prosthetics. The hippocampus seems to be fairly standardized hardware across mammals, so a black-box chip that could be shown to work on rats and primates might well work in the human brain. Additionally, determining the “codec” used to record memories seems to not only be possible, but universal enough that recording the firing patterns in one rat’s brain from when it learned a task and replaying them in another rat’s brain allowed the second rat to perform the task without training.

      This probably will not allow you or I to ever have our existing memories uploaded to any sort of computer, since this really takes place on the encoding end and not the readout end – but it may be possible to one day have our descendants record their experiences as their brain understands them – and possibly transfer them between brains.

  3. All I want to know is if these chips can be clustered into a group of twelve with an additional CPU, RAM, and IO preprocessor. I need a 15-chip SBC that is approximately 2.3 inches long, 1 inch wide, and .25 inches thick. Nothing else will fit properly into the CPU-slot of a T-800 unit.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.