Atlas Artificial Intelligence Computing Platform 

Huawei’s Atlas intelligent computing platform is based on Huawei’s Ascend series AI processors and the industry’s mainstream heterogeneous computing components. By a variety of product forms such as modules, boards, small stations, and AI servers ,huawei creates a “device, edge, and cloud to build a full-scenario AI infrastructure solution; Such solutions is widely used in fields such as “safe cities, smart transportation, smart medical care, and AI reasoning.”

Chinese name :Atlas Artificial Intelligence Computing Platform

Application areas :Safe cities, smart transportation, smart healthcare, AI reasoning, etc.

Developer: Huawei

Basic introduction to product series, product history, TA reference materials

On April 10, 2019, Huawei officially launched the Atlas artificial intelligence computing platform based on Ascend AI chips—a solution for all AI scenarios.

According to reports, the Atlas artificial intelligence computing platform includes Atlas 200 AI acceleration module, Atlas 300 AI accelerator card, Atlas 200 DK AI developer kit, Atlas 500 smart station, Atlas 800 AI server and other products.

However,These products can be applied to public security, operators, finance, Internet, power and other industries. For example, the Atlas 200 AI acceleration module can be used for cameras, drones and other terminals, and can support real-time analysis of 16 channels of high-definition video in half the size of a credit card.

Product Series including following:

1.Atlas 200 AI acceleration module

The size of half a credit card can support real-time analysis of 16 channels of high-definition video. It is deployed for end-side devices such as cameras and drones, as well as edge-side devices such as smart stations, with a TDP power consumption of 9.5W.

2.Atlas 300 AI accelerator card

It adopts the standard half-height and half-length PCIe card product form and is oriented to data center and edge server scenarios. The accelerator card supports a variety of data accuracies, and a single card provides 64 TOPS INT8 computing power and 64 channels of high-definition video real-time analysis. Provide more powerful computing power for deep learning and reasoning.

3.Atlas 500 smart station

It can achieve 16-channel high-definition video processing capabilities in the size of a set-top box, and its performance is 4 times higher than that of industry products. [8] Smart stations are suitable for a wide range of fields such as transportation, nursing, unmanned retail, and intelligent manufacturing.

4.Atlas 800 AI server

Based on standard servers and Atlas 300 AI accelerator cards, an integrated inference platform based on Kunpeng processors and x86 processors is built to provide real-time inference performance and high-efficiency computing features.

AI solutions

It Provides a series of software and hardware integrated delivery solutions and integrated optimization from deep learning [1], face recognition, vehicle recognition to text recognition.

1.Huawei Atlas 200 DK AI Developer Kit

It is a developer board product with the Ascend 310 chip as the core, helping developers of AI applications quickly become familiar with the development environment. Its main function is to open up the core functions of the Ascend 310 chip through the peripheral interface on the board, allowing users to quickly and easily access and use the powerful processing capabilities of the Ascend 310 chip.

2. Atlas 900 AI cluster

The Atlas900AI cluster represents the pinnacle of computing power. It is composed of thousands of Ascend 910AI processors. Through Huawei’s cluster communication library and job scheduling platform, it integrates three high-speed interfaces, HCCS, PCIe4.0 and 100GRoCE, to fully unleash the powerful performance of Ascend 910. Its total computing power reaches 256P~1024PFLOPS@FP16, which is equivalent to the computing power of 500,000 PCs. After actual testing, Atlas900 can complete training based on the Resnet-50 model in 60 seconds, which is 15% faster than the second place.

This allows researchers to conduct image and voice AI model training faster, allowing humans to more efficiently explore the mysteries of the universe, predict weather, explore for oil, and accelerate the commercialization of autonomous driving.

Product history

In September 2017, the first-generation Atlas intelligent cloud hardware platform was released at the Huawei HC2017 conference.

In March 2018, won the Grand Prize at Tokyo Interop Exhibition

In October 2018, the upgraded Atlas intelligent computing platform was released at the HUAWEI CONNECT 2018 conference. Based on Huawei’s self-developed Ascend chips, it provides full-scenario AI solutions.

In November 2018, won the China International Industrial Fair Award

In November 2018, Ascend 310 chip won the World Internet Leading Technology Achievement Award