The ‘surround computing’ era is coming soon, according to Mark Papermaster, the chief technology officer of chipmaker Advanced Micro Devices.
In a talk at the Hot Chips semiconductor design conference in Cupertino, Calif., Papermaster said that the enormous growth of sensors and mobile computing devices will produce a huge explosion of data that will overwhelm traditional data centers. We’ll be surrounded by computing everywhere, which explains why Papermaster calls the time ahead the surround computing era.
To deal with this surge of data, Papermaster said chip designers will have to create server chips with “heterogenouus” capabilities, acting as both microprocessor or graphics processor from one moment to the next. Sometimes you need a single processor (microprocessor computing) to work on a chunky piece of data, and sometimes you need a lot of processors working in parallel (graphics computing). AMD specializes in microprocessor-graphics combo chips dubbed accelerating processing units, or APUs.
“Of course, there is change in the industry and you can integrate more things together on a chip,” said Papermaster in an interview with VentureBeat before his talk. “But how you integrate things together matters. We’re going to have a data overload and an interface overload.”
Surround computing will put a lot of pressure on existing infrastructure. Unstrutured data such as video is expected to grow from 245 exabytes of data in 2010 to 1,000 exabytes by 2015, according to Cisco. Some 10 million new servers will likely be needed in data centers to deal with the change.
The technologies making the demands include natural user interface gesture recognition, fingerprint,or face recognition, augmented reality, content accessible anywhere, video game experiences, and audio-visual content management, he said. All of that computing will be done not at the point where the sensors are but in the mobile device clients and the centralized data centers, Papermaster said. Both clients and servers have to be smarter and far more power efficient.
More info here.