Step by step, artificial intelligence is moving down from the cloud and into the device in your hand. The latest sign? This unassuming little thumb drive from chipmaker Movidius, which packs one of the company’s machine vision processors — the same chip used by DJI for its autonomous drones — into a plug-and-play USB stick. If manufacturers want to beef up the AI capabilities of their new product, all they need to do is plug in one of these.
The Movidius Neural Compute Stick was actually announced last April as a prototype device called the Fathom. But then Intel came a’calling, and bought Movidius in September that year for an undisclosed amount. In all the work and confusion that comes with any sale like that, the Fathom got put on hold. Now though, it’s back.
From a technical point of view, the new Compute Stick is the same as the old one. At its heart is a Myriad 2 Vision Processing Unit or VPU — a low-power processor (it consumes just a single watt) that uses twelve parallel cores to run vision algorithms like object detection and facial recognition. Movidius says it delivers more than 100 gigaflops of performance, and can natively run neural networks built using the Caffe framework. (Caffe is one of the neural network libraries around, but it’s not clear if the Compute Stick will also work with Google’s popular TensorFlow framework.) For more details, you can check out the full spec sheet for the Myriad 2 here.
The main changes in this new version are that it’s made out of aluminum instead of plastic, and the price has been cut from a putative $99 for the original, to $79. Movidius says Intel’s involvement helped push this price down.
But who will use the Neural Compute Stick? Well, it’ll come in handy for a few different groups. AI researchers will be able to use the stick as an accelerator — plugging it in to their computers to get a little more local power when training and designing new neural nets. (Movidius notes that you can also chain multiple sticks together, boosting the performance linearly with each one you add). Companies looking to put AI powers in a physical product will also benefit, with the USB-compatible stick giving them an easy and fast way to execute neural networks locally.
But of course, a device like this certainly has its limitations. For a company building, say, an AI-powered security camera, there will be more efficient ways to incorporate specialized vision processors in their product, especially if they’re manufacturing at scale. And for a researcher training new neural nets, buying the latest graphic cards or renting processing power in the cloud will offer quicker results. It’ll just be more expensive too.
What a device like the Neural Compute Stick does well, is fill a gap in the market. And, in doing so, it make artificial intelligence that little bit more accessible.