There is a saying, for believers, that says that ‘God squeezes but does not drown’. If we see Huawei as the God of artificial intelligence processors, its latest AI chip may drown more than one competitor as it also comes with a super intelligent computing platform.
‘With nocturnal, premeditation and treachery’ is a phrase that is often used in criminal law to state that an act in question has been carried out with a series of aggravating factors such as those mentioned. This is what Huawei did, in this case without negative or criminal connotations of course, at 09:00 in the morning on Friday, August 23. With many of the journalists who talk about these things, especially artificial intelligence (AI), on vacation, and in my case without being able to connect to the streaming video of the rotating president of Huawei, Eric Xu, because in the mountainous area in the I’m not having enough coverage. But I did find out and read that in Shenzhen the Chinese firm announced the launch of a new AI processor and an intelligent computing platform that will become the most powerful of their generation, or at least that’s what they reported.
Ascend 910 is, according to Huawei, the most powerful AI processor in the world thanks to its 256 trillion floating point operations per second (256 TFLOPS) of medium precision, and 512 trillion operations per second (512 TOPS) of simple precision. But the most relevant thing is that with all the trillions of operations it performs per second it only consumes 310 watts of energy to unleash that computing power.
The announcement of Huawei’s new AI chip was accompanied by ‘MindSpore’, an artificial intelligence computing platform that supports the development of AI applications in any field so that they are also more widespread and accessible and guarantee protection and privacy. This new computing environment, according to the Chinese company, “is capable of supporting different situations at the same time, guaranteeing the security of private data, since it does not deal with the data itself but with gradients and processed models that no longer contain information. Of privacy. In addition to defending privacy, MindSpore has protection mechanisms built into the AI framework, making the models more secure and reliable. Based on versatility and synergy between different scenarios, including applications for the device, edge and Cloud, MindSpore operates on the concept of AI algorithm as code, in order to facilitate development and save the time needed to create models ” .
Using a natural language processing (NLP) network as an example, compared to other AI computing platforms, MindSpore can reduce the coding load by 20%, significantly lowering the access threshold for developers and improving overall efficiency. by more than 50%, and it is compatible with processing units such as GPUs and CPUs. According to Eric Xu, “MindSpore will be open source and will be enabled for developers from the first quarter of 2020, thus empowering the AI industry.”
The rotating President of Huawei also indicated that although AI is a new general-purpose technology, just as railroads and electricity were in the 19th century and automobiles and computers in the 20th century, its application is still in its infancy. initial phase and has a long way to go. “Our goal is to reduce, to eliminate, the gap between reality and expectation and accelerate the application of AI.”
And in this sense, Huawei intends to reduce with greater computing power the time required for training complex models from current days and weeks to minutes and seconds in the future; provide more affordable and abundant computing power to end scarce and expensive computing power, which limits AI development; enable the deployment of artificial intelligence in any scenario, not only in public clouds, adapting the portfolio to the diverse needs of companies and ensuring respect and protection of user privacy; Invest in the development of AI algorithms to optimize data efficiency consisting of lower data demand and energy efficiency, which is understood as lower computing power and energy consumption required; raise the level of artificial intelligence automation and minimize reliance on human intervention; constantly improve modeling algorithms to achieve excellent performance under real industrial conditions rather than good performance under virtual test conditions; implement real-time closed-loop systems to ensure optimal health for business AI applications; maximize the effect through the synergy between AI and other technologies such as 5G, Cloud, Internet of Things (IoT), edge, Blockchain, etc; build a platformfull-stack to make AI a basic skill for all developers and even all ICT professionals, and not just exclusively for specialized experts; and increase the availability of AI talent through technologiesfull-stackand multi-stage together with investment in talent development and an open ecosystem.
