Microsoft’s HPU 2.0 to power next-gen HoloLens devices

On Sunday, the Microsoft Research team announced they were developing a Holographic Processing Unit (HPU) along with an AI coprocessor to deal with Deep Neural Networks on the next generation of HoloLens devices.

Harry Shum, Executive VP of Microsoft’s AI and Research Group, revealed the HPU 2.0 to the crowd at CVPR 2017 in Honolulu, Hawaii. The idea behind the custom silicon is that the mixed reality headsets by the tech giant are able to process visual data within the device instead of relying on external servers.

With this move, Microsoft is entering uncharted territory and stepping into the turf of industry giants like Qualcomm and ARM. Leading semiconductor companies are also trying to develop self-contained, AI-processing units for mobile devices ahead of future necessities.

What are the implications and benefits of HPU 2.0?

In simple terms, Microsoft just wants to improve efficiency and user experience going forward. The solution devised to kill two birds with one stone is to go into the chip design business and develop a unit that can handle heavy workloads without decreasing performance and usage significantly.

This paradigm, of course, applies mostly to mobile devices of every kind, including the new lineup of affordable headsets pitched by Microsoft as pioneering pieces of hardware in the field of mixed reality.

The next generation of HoloLens devices is about to get more powerful to handle AI. Image: Microsoft.

However, HoloLens headsets will rely on an internet connection and external servers to process the gross of the visual data generated by users’ likely environments such as offices and workplaces in which they need real-time assistance as proposed by many in the industry.

As they are, the Microsoft headsets will be affordable enough for devices in their category, but the technology itself will be limited in contrast to its theoretical potential. The HPU 2.0 is a first step in the right direction to remedy this, and they will allow in-headset processing and even more wireless use than before.

Much like Google’s recently revived Glass initiative, the HoloLens lineup could serve as enterprise tools for workers looking for solutions in real-time with the help of enhanced reality. The necessary computing power to achieve this lies in deep neural networks, for which you would need an AI unit ready-to-go built in.

The same thing goes for museum installations based on augmented or virtual reality. To craft the experience to the users’ perspectives, machine learning models and processing need to be contained within the device to allow faster responses and more fidelity to the realities people are trying to create.

Microsoft’s HPU 2.0 is only being designed by the tech giant but will be built by someone else. Regardless, it will ship in HoloLens 2 headsets starting as early as 2019, according to sources familiar with the company.

Source: Microsoft