You may never have heard of Silicon Valley startup Movidius, but you’ve certainly heard about its unique vision processing chip, the Myriad VPU. It’s at the heart of Google’s Project Tango smartphone and tablet. Now, with its new, more-efficient version, the Myriad 2 VPU, Movidius is aiming to bring the high-end vision and computational imaging applications that have only been possible using power-hungry, computer-based, GPU solutions to mobile devices. The Myriad 2 is designed to replace or at least augment the custom silicon and mobile GPU solutions currently used for similar applications.
Myriad 2: 20x improvement over the Project Tango chip
For anyone who has seen Project Tango in action, it is already amazing that it can do real-time 3D vision and mapping on a tablet, helped along by the original Myriad chip paired with an Nvidia K1 GPU. Movidius claims its new Myriad 2 chip is 20x more efficient than than that version. For starters, it should make it possible to shrink the Tango tablet into smaller form factors.
The Myriad 2 VPU (vision processing unit) can execute over 2 trillion 16-bit operations per second while consuming only 500 milliwatts. That makes it ideal for the massive data throughput requirements of real-time vision and video processing applications. Like a GPU, the Myriad 2 is programmable, with 12 vision-specific vector processors that can either work as coprocessors, or even as standalone processors in dedicated devices. Movidius provides a graphical development environment that supports both C++ and OpenCL.
Transforming photography: The promise of computational imaging
Movidius CEO Remi El-Ouazzane sees computational imaging as the third wave of photography, using computational power to radically improve photography by adding features like depth maps, faster autofocus, realistic panoramas, and integrated HDR processing. In this he joins other visionaries including Lytro’s Ren Ng and Pelican’s Kartik Venkataraman. Unlike those companies, Movidius isn’t building the actual imaging devices, instead aiming to embed its highly-specialized SoCs in a variety of OEM solutions. Movidius claims that using its technology adds less then $10 to the cost of a mobile device. For comparison, the camera module in a high-end smartphone is estimated to cost around $20, so adding a VPU would increase the cost of the camera by about 50%.
Much in the same way that off-the-shelf GPUs rapidly gained market share over homebrew, custom-silicon solutions to accelerated graphics, VPUs provide a compelling value proposition to anyone needing either computational photography or vision solutions in a mobile device. There is no doubt that a few dollars per unit is not a high price to pay to get both a high-performance tried-and-tested VPU, and a full set of development tools. For now, Movidius has this space to itself, leaving it with the challenge of gaining mind and market share before larger competitors jump into the space.
©TechSenze
