Microsoft designed its own AI chip for the next HoloLens


There's a lot of focus right now on the future of virtual reality headsets, the cost of entry, and just how good our VR experiences could get.

The second version of Microsoft's HoloLens, the firm's "mixed-reality" headset, will feature an on-board AI coprocessor designed by Microsoft itself, Microsoft's AI lead Harry Shum announced in a keynote address at CVPR 2017 on Sunday. The other approach to unlocking future potential is to create lightweight neural networks which remove the strain from device processors as Google and Facebook are doing. The new AI coprocessor runs off the HoloLens 2 battery.

The current iteration of the HoloLens is now only available to developers and is still waiting on a commercial release. Sadly, the company's initially available HoloLens hardware is priced out of the reach of most consumers, costing $3,000 for the Development Edition release - a price that is being brought down by hardware partners working on third-party lower-specification headsets based on the same Windows Mixed Reality platform.

The second version of Hololens HPU will incorporate an AI coprocessor to natively and flexibly implement Deep Neural Networks (DNNs). This also means that user data is more secure as it remains on the device at all times.

HoloLens contains a custom multiprocessor called the Holographic Processing Unit, or HPU.

Pollefeys goes on to give additional background on the HPU, with all the tech and sensors that go in to making the HPU, which inside HoloLens makes it "the world's first-and still only-fully self-contained holographic computer".

We all know that Microsoft has been now working on launching the consumer ready version Hololens, the standalone mixed reality headset. With this chip, HoloLens 2 will be able to analyse visual data directly, without having to upload the data to the cloud, which should result in a quicker, more efficient performance. The AI coprocessor can run continuously, off the HoloLens battery, allowing features such as sophisticated hand tracking with hand-part segmentation. One of its closest rivals is Magic Leap.