2019 Annual Report (^47)
■ Ascend 310 AI processor for inference: The Ascend
310-powered Atlas series accelerator module,
accelerator card, edge station, and server are now
part of dozens of industry solutions (e.g., smart
transportation, smart grid, smart finance, smart city,
and smart manufacturing) developed by dozens
of partners. Our mobile data center (MDC) – a
computing platform for intelligent driving based
on the Ascend 310 processor – has been used by
18 leading automakers and integrators to provide
intelligent driving solutions across a variety of
scenarios, including passenger vehicles, commercial
vehicles, and remotely controlled vehicles.
■ Ascend 910 AI processor for training: For
half-precision floating point (FP16) operations,
Ascend 910 delivers 256 TFLOPS. For integer
precision calculations (INT8), it delivers 512 TOPS.
Its maximum power consumption is only 310 W.
■ Compute Architecture for Neural Networks (CANN):
The efficient programming and execution framework
based on the Da Vinci architecture that provides a
series of application programming interfaces (APIs)
and tools for operator development and network
optimization. It helps build open, efficient, and
high-performance development environments across
layers and domains for different types of developers,
which helps make hardware more effective and
easier to use.
■ MindSpore: The all-scenario AI computing
framework that can adapt to all scenarios – across
devices, edge, and cloud – and provides a secure
and trustworthy inference and training framework
for consistent development and on-demand
cooperation. The innovative “AI Algorithm as
Code” design makes developing AI applications
simpler and lowers the threshold for model
development. Through technological innovation
and a co-optimization of MindSpore and Ascend,
our solution ensures runtime efficiency and helps
developers address complex AI computing challenges
and their need for diversified computing power.
■ ModelArts: The one-stop AI development and
management platform that provides model
development services spanning the full pipeline,
from data access and model development to model
training and deployment.
■ Atlas-powered AI computing products: We have
rolled out a full Atlas portfolio including the Atlas
200 Al Accelerator Module, Atlas 300 AI Accelerator
Card, Atlas 500 AI Edge Station, Atlas 800 AI
Server, and Atlas 900 AI Cluster. Our portfolio
provides powerful computing capabilities for AI
training and inference across device, edge, and
cloud scenarios.
■ Ascend-powered AI cloud services: HUAWEI CLOUD
has rolled out 43 AI cloud services powered by
Ascend in order to unleash computing power. Our
AI cloud servers improve performance by more
than twofold and are widely used in AI inference,
AI training, and autonomous driving training.
Additionally, our industry’s first enterprise-class
Knowledge Graph makes construction of knowledge
graphs over 70% more efficient.
Forging ahead with a computing strategy for a
thriving computing industry: “one cloud” + “two
wings” + “two engines”
In September 2019, Huawei announced its computing
strategy. Driven by this strategy, we aim to provide
the ultimate computing power to deliver ubiquitous
cloud and pervasive intelligence. If the entire
computing ecosystem were a towering tree, then
Kunpeng and Ascend would be the roots. It’s only
after the roots reach deep into the soil that the tree
becomes sturdy. To strengthen the computing industry,
Huawei harnesses the power of “one cloud” (HUAWEI
CLOUD), “two wings” (computing and intelligent data
& storage), and “two engines” (Kunpeng and Ascend).
Specifically, we continue to pursue innovation in the
following areas:
■ Cloud: Our solutions enable greater synergy
between cloud, AI, and 5G. This synergy, coupled
with full-stack technological innovation, translates
into several layers of added value in our public
cloud services and hybrid cloud solutions,
especially in terms of stability, reliability, security,
trustworthiness, and sustainability. We are
working with our partners to help industries more
effectively navigate their digital and intelligent
transformations. Our goal is to better support
our partners’ applications and help customers do
more with their data, creating fertile soil for the
intelligent world to flourish.
■ Computing power: For the past 15 years we have
been investing in the development of a chipset
family built with Kunpeng and Ascend at the
core. We have developed integrated equipment,
boards, and cards for general-purpose computing
scenarios. For AI computing, we have the full series
of Atlas products, supporting devices, edge, and
data centers.
coco
(coco)
#1