Huawei Launches Ascend 910 AI Processor and MindSpore Dev Framework
- By John K. Waters
Huawei Technologies has announced the commercial availability of the Ascend 910 chip, the long-awaited AI processor aimed at datacenters and designed to compete with top U.S. chip makers, such as Intel, Qualcomm, Samsung, and Nvidia. The company also launched a new AI computing framework, called MindSpore. Huawei launched the products despite currently being prohibited from doing business with U.S. companies.
Originally unveiled last October, the 7nm Ascend 910 was designed for AI model training. In a typical training session based on the ResNet-50 neural network, Huawei claims, the combination of Ascend 910 and the MindSpore framework is about two times faster at training AI models than other mainstream training cards using TensorFlow. For half-precision floating point (FP16) operations, Ascend 910 delivers 256 TeraFLOPS, the company said. For integer precision calculations (INT8), it delivers 512 TeraOPS. The chip's maximum power consumption is 310W, which the company noted is lower than its projected 350W.
The chip belongs to Huawei's Ascend-Max family of chipsets, along with the previously launched Ascend 310.
Huawei calls its new MindSpore AI computing framework an "all-scenarios" framework that can adapt to deployment needs across all devices, edge computing, and cloud environments, and provides on-demand cooperation among them. The company characterizes this capability as "AI Algorithm as Code" that improves simplicity and cuts development time. In a typical neural network for natural language processing (NLP), for example, MindSpore has 20 percent fewer lines of core code than leading frameworks, and it helps developers raise their efficiency by about 50 percent, the company said. MindSpore has been optimized for the new Ascend processors, but also supports CPUs, GPUs and "other types of processors."
"We have been making steady progress since we announced our AI strategy in October last year," said Eric Xu, Huawei's current Rotating Chairman, in a statement. "Everything is moving forward according to plan, from R&D to product launch. We promised a full-stack, all-scenario AI portfolio. And today we delivered, with the release of Ascend 910 and MindSpore."
Xu added that this launch marked "a new stage in Huawei's AI strategy."
Huawei's published AI strategy involves investing heavily in research, developing fundamental machine learning capabilities in computer vision, NLP, and decision and inference, and then building a full-stack AI portfolio. Huawei views AI as "a new general-purpose technology, like railroads and electricity in the 19th century, and cars, computers, and the Internet in the 20th century. The company believes that AI will be used in almost every sector of the economy." The company unveiled its first AI chip, the Kirin 970, in 2017. The Kirin 980 followed a year later.
Huawei is currently on a U.S. blacklist known as The Entity List, which restricts American companies from doing business with the company. It's also being investigated by U.S. prosecutors for alleged theft of intellectual property, The Wall Street Journal reported. Huawei is accused of stealing intellectual property from individuals and companies over several years, and recruiting employees from its rivals, the newspaper reported.
Being blacklisted means Huawei's latest AI offering is off limits to American companies. Google has said that those restrictions will prevent Huawei's upcoming 5G-capable Mate 30 line of smartphones from using Google apps and services, including Maps, YouTube, Pay, Photos or the Play Store.
About the Author
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at firstname.lastname@example.org.