Huawei Atlas 900 AI Cluster Wins the Only Best of Show Award for AI at Interop Tokyo 2020 – PRNewswire

Tony Xu, President of Huawei Ascend Computing, said, "The Huawei Atlas AI computing solution provides powerful computing and ultimate energy efficiency for all AI scenarios across devices, the edge, and the cloud. The Atlas 900 AI cluster provides data centers with powerful computing, high linearity, and the best energy efficiency to accelerate data-intensive research, such as astronomical exploration, weather prediction, oil exploration, and gene sequencing. Research breakthroughs translate into practical benefits for people worldwide."

Atlas 900 is the fastest AI training cluster in the world. It delivers a total computing power of 256 to 1024 petaFLOPS at half precision (FP16), equivalent to the computing power of 500,000 personal computers. Atlas 900 shattered the world record on the ResNet-50 benchmark test for model training by completing training in 59.8 seconds. Atlas 900 is the only product capable of completing the training in under a minute. Atlas 900 has broad applications in scientific research and business innovation for faster training of AI models with images and videos.

Atlas 900 integrates three interfaces for high-speed interconnection: Huawei Cache Coherence System (HCCS), PCIe 4.0, and 100G Ethernet. The Atlas 900 AI cluster leverages the Huawei CloudEngine data center switches to work on a 100 TB/s full-mesh, non-blocking dedicated network for parameter synchronization. The network slashes parameter synchronization latency by 10 to 70 percent to streamline AI model training.

Heat dissipation is a critical issue for an AI training cluster with such high computing power. That is why the Atlas 900 AI cluster adopts a groundbreaking system for heat dissipation. It leads industry innovation with a full liquid cooling solution and a rack-scale enclosed adiabatic design. This design delivers tremendous heat dissipation even for single racks with power consumption of up to 50 kW. It achieves a power usage effectiveness (PUE) of below 1.1 for data centers, almost reaching the ideal PUE of 1.0. Atlas 900 improves over air-cooled 8-kW racks by reducing equipment room space by 79%. Its innovative liquid cooling system provides energy-intensive, high-density, and low-PUE deployment to drastically reduce customer TCO.

Huawei is fostering cooperation to build the Ascend computing industry with open hardware, open source software, and partner enablement. Huawei provides full-stack AI computing infrastructure and application solutions to power industries with AI and create pervasive intelligence.

--Ends--

About Huawei

Huawei is a leading global provider of information and communications technology (ICT) infrastructure and smart devices. With integrated solutions across four key domains telecom networks, IT, smart devices, and cloud services we are committed to bringing digital to every person, home and organization for a fully connected, intelligent world.

Huawei's end-to-end portfolio of products, solutions and services are both competitive and secure. Through open collaboration with ecosystem partners, we create lasting value for our customers, working to empower people, enrich home life, and inspire innovation in organizations of all shapes and sizes.

At Huawei, innovation focuses on customer needs. We invest heavily in basic research, concentrating on technological breakthroughs that drive the world forward. We have more than 194,000 employees, and we operate in more than 170 countries and regions. Founded in 1987, Huawei is a private company wholly owned by its employees. For more information, please visit Huawei online at http://www.huawei.com or follow us on:

http://www.linkedin.com/company/Huawei http://www.twitter.com/Huawei http://www.facebook.com/Huawei http://www.youtube.com/Huawei

Photo - https://mma.prnewswire.com/media/1176654/Huawei.jpg

http://www.huawei.com

SOURCE Huawei

Follow this link:
Huawei Atlas 900 AI Cluster Wins the Only Best of Show Award for AI at Interop Tokyo 2020 - PRNewswire