Huawei Adds Ascend-based Atlas Series Products and Cloud Services to Its Full-Stack All-Scenario AI Solutions
[Shanghai, China, September 19, 2019] At this year's annual HUAWEI CONNECT, Huawei launched its Atlas series of products and 43 cloud services based on Ascend processors with powerful computing capabilities. The Atlas 900 AI cluster, Atlas 300 AI training card, and Atlas 800 AI training server are among the products to have attained the pinnacle of computing power across the globe. Thanks to continuous innovation of Ascend processors, the leading ICT solutions provider offers customers more abundant, economical, and adaptive AI algorithms for all scenarios. This launch further enriches Huawei's portfolio of full-stack all-scenario AI solutions.
"We're in the golden age of computing architecture. Abundant, high-performance, diversified, greener and inclusive computing power will drive the development of an intelligent society. Our goal is to innovate to customer demand and provide computing power for ubiquitous cloud and pervasive intelligence. Our Ascend-based Atlas series of products and cloud services are part of Huawei's AI strategy. The powerful computing capabilities of Ascend processors are now available through cloud and hardware, making computing power accessible and AI inclusive." Hou Jinlong, President of Huawei's Cloud & AI Products and Services, said in his keynote speech titled Empower New Computing with Ascend + Kunpeng.
Hou Jinlong, President of Huawei's Cloud & AI Products & Services, delivers his keynote speech
43 Ascend-based New Cloud Services
The services launched are built on Ascend 910 and Ascend 310 and tailored to scenarios such as AI data inference, AI data training, image processing, knowledge computing, and autonomous driving training.
Thanks to continuous innovation, HUAWEI CLOUD now offers high-performance and cost-effective AI services to a wide range of industries for inclusive AI for all.
AI cloud servers are now doubly powerful. With Huawei's enterprise-class knowledge graph, petroleum industry experts use multi-source knowledge convergence technologies to become over 70% more efficient. Autonomous driving in the cloud uses ultra-large clusters to process PBs of data in real time, making E2E development over five times more efficient.
Atlas Series: the World's Most Powerful Computing Products
Built on Ascend 910 and 310 AI processors, the world's fastest AI training cluster Atlas 900, AI training server Atlas 800, AI edge station Atlas 500, AI inference and training card Atlas 300 as well as AI accelerator module Atlas 200 are part of Huawei's complete Atlas product portfolio. The portfolio covers all cloud, edge, and device scenarios and offers the computing power needed for training and inference. Featuring a unified Da Vinci architecture and all-scenario AI computing framework, the Atlas series of products leverages cloud-edge-device synergy to accelerate intelligent industry reconstruction.
Michael Ma, President of Huawei's Intelligent Computing Business Dept, launches Atlas 300 and Atlas 800
- Atlas 900: This scalable AI cluster consists of thousands of Ascend 910 processors. The Atlas 900 ranked No. 1 in the world in ResNet-50 testing with a 59.8-second duration, 15% faster than the runner-up. Atlas 900 possesses the computing power for scientific research, space exploration, and oil exploration.
- Atlas 800: This 4U server integrates eight Ascend 910 AI processors for 2 PFLOPS of computing power, at a density 2.5 times that of the competition. The Atlas 800 is only 75 kg, less than half of the weight of the competition. Air and liquid cooling meets high-density deployment requirements of enterprise data centers and clusters. Each server has 1.8 times the energy efficiency ratio of the competition.
- Atlas 300: This AI training card provides 256 TFLOPS of computing power for an increase of images trained per second from the mainstream 965 to 1802, or twice that of the industry's most powerful training cards. The Atlas 300 uses pass-through high-speed 100G RoCE interfaces for parallel transmission of gradient parameters and datasets, reducing synchronization delay by up to 70% and shortening cluster training time to just seconds.
Up to now, Huawei has launched Ascend AI processors, Ascend-based Atlas series products and cloud services, and all-scenario AI computing framework MindSpore. Together, these innovations make up full-stack all-scenario AI solutions.
Huawei's next steps are further implementation of Ascend AI processors in its AI strategy, and further partnerships to promote AI for everyone.
An Open, Win-Win Ecosystem with Ascend and Kunpeng
Beyond Ascend processors, Huawei joins hands with partners to build the Kunpeng ecosystem. This full-stack infrastructure covers PC, server, storage, operating system, middleware, virtualization, database, cloud services, and applications. Huawei also launched 69 Kunpeng-based cloud services and hardware boards to accelerate customer innovation. Guided by Huawei’s Kunpeng board and reference design, partners can develop servers and desktop products rapidly. To promote this development, OpenGCC, China Electronics Standardization Institute, Arm China, IDC, Huawei, and industry partners launched the White Paper on Kunpeng Computing Industry Development. For details, please visit http://opengcc.org/forums/194.
White paper launch ceremony
The server operating system and database are essential for enterprise IT systems. Huawei has announced its open-source openEuler server OS available on December 31, 2020, and openGauss database available on June, 2020.
In this era of diversified computing, Huawei works with industry partners to build a computing ecosystem built on Ascend and Kunpeng processors. This ecosystem will provide leading IT infrastructure and applications for all industries.