Thursday, July 31, 2014

Movidius Announces its 2nd Gen Vision Processor

Marketwired: Movidius announces Myriad 2, the next generation of its vision processor unit (VPU) SoC. Compared to Myriad 1—which was introduced as part of Google’s Project Tango—the new chipset delivers 20x more processing efficiency in terms of computations per watt of power consumed. Myriad 2 comprises a larger and upgraded set of programmable processors as well as a new set of dedicated and configurable image and vision accelerators to power the next wave of computational cameras.

Its key features:
  • Superior performance at ultra-low power: Myriad 2, implemented in TSMC 28HPM process, is capable of performance exceeding two trillion 16-bit operations per second while consuming an average of less than 500mW. It supports up to six full HD 60fps camera inputs simultaneously via 12 1.5Gbps MIPI lanes.
  • Programmability: Myriad 2 is equipped with 12 programmable vision-specific vector processors to enable rapid innovation and to enable device manufacturers to differentiate using their proprietary software.
  • Low latency processing: Given its highly parallelized data processing architecture and its new and highly efficient on-chip memory fabric, Myriad 2 can achieve high-performance processing with notably low latency. As an example, for the standard vision processing benchmark known as the Haar Cascade classification, Myriad 2 can calculate 50,000 multi-scale classifications per high-definition (HD) video frame in a mere seven milliseconds.
  • Flexible usage for different applications: Myriad 2 operates either as a co-processor to the main application processor in a vision-enabled mobile device, or as a standalone processor in wearable or embedded applications.

1 comment:

  1. What hardware accelerators are included? .5W is very impressive for what they can achieve. I wonder what all applications will use this new chip? Amazon Fire Phone version 2 perhaps? Maybe next-gen LG and Samsung smartphones?

    Does anybody know if Apple is adding similar hardware to process the PrimeSense depth data?


All comments are moderated to avoid spam and personal attacks.