» Current | 2017 | 2016 | 2015 | Subscribe

Linley Newsletter

Synopsys EV6x Serves Up Big MACs

August 8, 2017

Author: Mike Demler

Synopsys has upgraded its DesignWare EV6x family of embedded-vision processor cores, adding new configurations that enable designers to include up to four convolutional-neural-network (CNN) accelerators. Together, these accelerators comprise a massive 3,520 single-cycle multiply-accumulators (MACs). When manufactured in a 16nm FinFET process, the new EV6x can clock at up to 1.28GHz, delivering peak CNN performance of 4,500 GMAC operations per second (GMAC/s). The company plans to release the latest EV6x intellectual property (IP) for general licensing by the end of August.

Whereas the EV6x models that Synopsys unveiled last year optionally include a single 880-MAC accelerator, the upgrades allow designers to integrate one, two, or four of the convolution and classification engines. The maximum configuration offers CNN acceleration similar to that of the dual 2,048-MAC deep-learning accelerators in Nvidia’s new Xavier SoC. That product delivers 5,120GMAC/s to power Level 4/5 autonomous-driving systems.

As in the initial design, the new EV6x uses 12-bit integer MACs to offer a middle ground between the typical 8- and 16-bit units in other computer-vision (CV) cores. The 12-bit MACs consume only half the silicon area of 16-bit MACs, and they yield higher accuracy than 8-bit integer CV cores.

The scalability of the EV6x in CPUs/DSPs and CNN accelerators enables licensees to tune it for a wide range of embedded-vision tasks. Although designers who can accept the lower precision of an 8-bit CNN may prefer a core built for that resolution, the upgraded EV6x architecture offers performance that matches the best licensable CV IP.

Subscribers can view the full article in the Microprocessor Report.

Subscribe to the Microprocessor Report and always get the full story!

Purchase the full article

Events

Linley Processor Conference 2017
Covers processors and IP cores used in deep learning, embedded, communications, automotive, IoT, and server designs.
October 4 - 5, 2017
Hyatt Regency, Santa Clara, CA
Register Now!
More Events »

Newsletter

Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »