» Current | 2020 | 2019 | 2018 | Subscribe

Linley Newsletter

Huawei Ascends in AI Accelerators

May 14, 2019

Author: Linley Gwennap

Aiming for world leadership in AI, Huawei rolled out a set of neural-network accelerators including the Atlas 200 module for smart cameras and the Atlas 300 card for edge compute. These products use the Ascend 310, a chip the company designed itself. It employs an architecture called DaVinci that natively implements 3D tensor computations. The company is developing a more powerful chip, the Ascend 910, but products using this device have been delayed until later this year.

The DaVinci core contains what Huawei calls a 3D Elastic Cube comprising 16x16x16 multiply-accumulate (MAC) units, or 4,096 total units. Each unit can perform one FP16 MAC or two INT8 MACs per cycle. Every DaVinci core is an independent programmable processor that contains a scalar ALU, a vector ALU, a load/store unit, and the tensor cube. Also in each core is an unspecified amount of memory that can serve as cache or SRAM.

A single chip can have one or more DaVinci cores along with one or more Arm CPUs. The Ascend 310 is a complete SoC that delivers 8 FP16 teraflops per second (Tflop/s) or 16 INT8 TOPS. At this performance, it consumes a maximum of 8W. Including the Arm CPU and other components on the chip, the efficiency works out to 2 TOPS per watt. The 310 can decode 16 video channels or encode 1 channel, assuming 1080p video in H.264 or H.265 format. Huawei also provides a set of drivers that support leading AI frameworks such as Pytorch and TensorFlow as well as its proprietary MindSpore framework.

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


Linley Spring Processor Conference 2020
Coming April 7-8, 2020
Hyatt Regency, Santa Clara, CA
Register Now!
Linley Fall Processor Conference 2020
Coming October 28-29, 2020
Hyatt Regency, Santa Clara, CA
More Events »


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