movidius_fathom_02Intel is acquiring Movidius, a chip-maker known for the Myriad series of SoC specialized in vision and deep learning inference workload at low power. Intel’s main stated reason for buying Movidius is to strengthen its Intel Realsense technology which relies on enhanced webcams (color+depth) to let computers know what their surrounding is like.

For Movidius, the acquisition comes relatively quickly since the company has been in existence only a few years, with its first chip the Myriad 1 being released on February 21, 2014, as part of the Google Tango launch. At that time, Remi Remi El-Ouazzane, the CEO of Movidius, had already told me that the end game was “to give human-like vision to your handset.”

movidius_fathom_04The Myriad 2 SoC was announced in July 2014 to bring even higher performance per Watt and accelerate the most common use cases by hardwiring some operations, while Myriad-1 only used programmable units. The additional hardware increases speed and reduces power usage by streamlining computations with dedicated computation units. Movidius says that Myriad 2 has an aggregate computation power of 2 Teraflops.

On April 2016, Movidius capitalizes on the rise of Deep Learning AI and computer vision by adding neural-net inference to the list of what its chip can do very well, and announces the Movidius Phantom, a USB computer based on the Myriad 2 chip. With it, “dumb” drones could become “smart” with the ability to recognize objects they see – all of this under a 1 Watt power budget.

Movidius chips’ advantages over Intel’s CPUs and NVIDIA/AMD GPUs is their low-power nature. On one hand, it is parallelized enough that a standard CPU cannot reach its power efficiency. On the other hand, it is much better than GPUs at branch execution, so GPUs have a hard time competing within the same power budget.

This built-in advantage designed into Myriad from the get go pays off today with Intel recognizing that it is better off buying Movidius to gain expertise and save time. The deep learning and computer vision train has left the station and Intel wants to be ahead in the race amid extremely stiff competition. This may prove to be a critical acquisition.

Filed in Computers. Read more about Intel, processors and SoC.

User Comments