2019 Annual Report (^49)
Leading diversified computing and developing
the industry ecosystem
Huawei hopes to drive the computing industry forward
with diversified computing across devices, edge, and
cloud with Kunpeng, Ascend, x86, and GPUs. To this
end, we have continued investing and innovating in
general-purpose computing and AI computing, in order
to provide the ultimate computing power.
Among these products is the Atlas 900 – the
world’s fastest AI training cluster, composed of
thousands of Ascend 910 AI processors. Atlas
900 delivers up to 256–1,024 PFLOPS at FP16,
equivalent to the computing power of 500,000
PCs. Tests conducted based on the most typical
ResNet-50 v1.5 model and ImageNet-1k dataset
showed that it takes Atlas 900 just 59.8 seconds
to complete the training, making it the fastest
training cluster in the world. Researchers can use
the leading computing capacity of Atlas 900 to
more rapidly train AI models for image and speech
recognition. Our AI cluster can also be used to
more efficiently forecast weather, search for oil,
accelerate the commercialization of autonomous
driving, and uncover the mysteries of the universe.
Atlas 900, the world’s fastest AI training cluster
■ General-purpose computing: Huawei has worked
with a variety of partners to develop the computing
industry and build full-stack IT infrastructure and
industry applications. The TaiShan series servers
are the result of the extensive expertise Huawei
has gained in computing technology and integrated
equipment engineering over the years. They
have already been extensively deployed in many
application scenarios such as big data, software-
defined storage, databases, web applications, cloud
phones, and high-performance computing.
■ AI computing: Huawei released its Atlas series
products powered by Ascend AI processors. These
products include the Atlas accelerator module,
accelerator card, edge station, server, and cluster.
Coupled with MindSpore and open development
tools such as CANN, we provide all-scenario (i.e.,
devices, edge, and cloud) AI infrastructure that
spans the entire deep learning process from AI
inference to training.
Based on these Atlas series products, Huawei is
working with over 100 independent software and
hardware vendors worldwide to use AI to empower
numerous industries and accelerate development
towards the intelligent world. In addition, Huawei
has worked with over 30 universities around
the world to develop educational courses and
build research labs, and helped more than 100
universities teach hands-on AI skills. Together, we
have achieved many impressive innovation results
in areas such as robotics, machine vision, and
image processing, promoted the development of
academic communities, and cultivated talent for
the AI industry.
Maximizing the value per bit and reducing cost
per bit to unlock the full potential of data
To accommodate massive data growth and the
rising demand for intelligent services, enterprise IT
must shift its current focus on storage towards data
infrastructure. Huawei provides converged, intelligent,
and open data infrastructure solutions that help
customers fully unleash data value and embed AI
into all aspects of operations. These solutions were
With an intelligent architecture based on “three flows”
and “one cloud”, our TaiShan server-based intelligent
manufacturing solution helps equip production sites with
ICT capabilities, such as Atlas-powered deep learning and
intelligent vision. Our solution allows customers to achieve
highly automated pull production and one-piece flow,
ensuring product delivery quality and meeting requirements
for mass customization.
“Three flows”: product engineering data flow, business
information flow, and production process flow
“One cloud”: cloud-based Manufacturing Execution System
(MES) + manufacturing IT system
coco
(coco)
#1