PantoInspect launches new digital twin platform for pantograph condition monitoring
Grant for grommets
Pantonium Launches Largest On-demand Transit Pilot in Sioux Falls using Full-sized Buses
Network Rail Grants Full Acceptance for CableGuardian®
Loram subsidiary GREX now known as Loram Technologies
In-Depth Focus: A Digital Future for Rail Freight
A sustainable solution to mitigate ground-borne vibration caused by trains
Discover how your workforce can be fully optimised
Mathematical framework helps to shape any sheet of material
Improving information for passengers and operators
Having announced the processor’s planned specifications almost a year ago at Huawei Connect 2018, Huawei officially launched the Ascend 910 processor and an “all-scenario AI computing framework”, MindSpore. The Ascend 910 is an AI model training processor belonging to Huawei’s Ascend-Max series of chipsets, and is said to fulfil or exceed all planned performance goals.
Eric Xu, Huawei’s rotating chairman, presented the new additions to the company’s full-stack AI portfolio in a livestreamed presentation from the company’s headquarters in Shenzhen, China, commenting: “We promised a full-stack, all-scenario AI portfolio. And today we delivered, with the release of Ascend 910 and MindSpore. This also marks a new stage in Huawei’s AI strategy.”
Despite its record-beating performance, the Ascend 910 has a maximum power consumption of just 310W, compared with a target of 350W. Huawei stated that the combination of Ascend 91 and MindSpore allows for training AI models approximately twice as fast as rival products, using open-source machine learning library TensorFlow.
“Ascend 910 performs much better than we expected,” Xu said. “Without a doubt, it has more computing power than any other AI processor in the world.”
The MindSpore framework is adaptable to all sorts of scenarios, including device, edge, and cloud applications (as well as combinations of these scenarios), and to all budgets. In a typical neural network for natural language processing, MindSpore has 20 per cent fewer lines of core code than leading frameworks on the market, reportedly helping developers increase efficiency by at least 50 per cent.
The new processor will be available in China from September and will go open-source in the first quarter of 2020 in a move to “drive broader AI adoption”.
According to Xu, Huawei believes that AI will eventually become a true general-purpose technology – like the internet has become – although it is still in early stages of development. He said that Huawei wished to drive much of its development through its AI strategy, including by investing in efficient basic AI algorithms, using MindSpore to help automate AI development, and providing more affordable and stronger computer power.
Huawei, which is the world’s largest telecommunications manufacturer, has been caught up in a trade war between the US and China. In May, the company was added to the US Entity List by executive order, preventing US companies from working with it without a special permit. Although this week, US authorities have announced that Huawei’s initial 90-day grace period – in which it is permitted to continue to buy from US suppliers – will be extended, uncertainty remains over its future work with US companies and allies.
Addressing the impact of its blacklisting at the press conference, Xu commented that the company was “fully prepared” to work under the US trade restrictions. He added that the company was experiencing less of an impact on its business than it had feared; Huawei CEO Ren Zhengfei had predicted a $30bn blow to its revenue.
“It seems it’s going to be a little less than [$30bn], but you’ll have to wait until our results in March,” Xu said.
This article first appeared on eandt.theiet.org
About this website
Railpage version 3.10.0.0037
All logos and trademarks in this site are property of their respective owner. The comments are property of their posters, all the rest is © 2003-2021 Interactive Omnimedia Pty Ltd.
You can syndicate our news using one of the RSS feeds.