HUAWEI CONNECT 2019: Computing Power is the Cornerstone in the AI Age
In the imminent future of advanced AI, China’s high-tech poster boy Huawei is now squarely training its guns on becoming a super computing powerhouse.
Although it wasn’t mentioned outright, it was clear to all at the HUAWEI CONNECT 2019 keynote address delivered by Deputy Chairman Ken Hu , that that the company has set its sights to be in the race to supply super computing power to the world, in abundance.
“For the past 30 years, we have invested non-stop in Connectivity. When people think Huawei, they think connections. However to build an intelligent world, Connectivity and Computing are inseparable. Where there is Connectivity, there must be is Computing.
Over the last decade, Huawei has invested heavily in Computing. Today moving into a world of Advance Intelligence, both (connectivity and computing) pushes the other forward, both developing step-in-step,” said Hu in addressing the thousands of attendees in audience at the Shanghai World Expo.
Hu quotes Gartner that by 2023, the Computing market will be worth over a staggering USD2 Trillion US, thus promising a massive ‘blue ocean market’. This, he said, calls for Huawei to have outlined four strategic areas to pour their R&D and investments into:- Architectural Innovation, AI-scenario processors, Business strategy and Open Ecosystem
ALTAS 900 – The World’s Fastest AI Training Cluster
Huawei’s biggest Computing announcement at Huawei CONNECT 2019 is the release of the Atlas 900, which it claims to be the world’s fastest ‘AI training cluster’. This powerhouse of AI computing, is to make AI more readily available for different fields of scientific research and business innovation.
Hu demonstrated to an audibly impressed audience how the Atlas 900 only required only 59.8 seconds – a significant 10 seconds faster than the previous world record, to train ResNet-50.
(Editor’s Note: Resnet-50 is the neural/ deep residual learning networks which led to breakthroughs in image classification. It is widely accepted as the industry’s ‘gold’ standard for measuring AI training performance.)
“Atlas 900 will bring new possibilities to different fields of scientific research and business innovation – anything from astronomy, weather forecasting, and autonomous driving, to oil exploration,” shared Hu.
In another ‘live’ demo, Altlas 900 showed how it took just a fraction of the time to do what conventionally took several months to train. The demo video showed the mapping of over 200,000 star in the Southern Hemisphere; whereby if an astronomer wanted to find a celestial body with specific features, it would take 169 days of full-time work with the existing computing capability. And the Atlas 900? It only took 10:28 seconds to scan through mountains of data to locate and identify a specific type of star.
“From 169 days to 10 seconds. It’s nothing less than revolutionary. And what more, now Huawei has also deployed Atlas 900 on Huawei Cloud as a cluster service so ask to make sheer computing power more accessible across different industries.”
It was also shared that Huawei is currently also involved to build China’s first evolving AI supercomputing system, using the Atlas 900 that supports exascale computing. This is in collaboration with the Peng Cheng Lab that focuses to establish a new generation of platforms for AI basic research and innovation.
AI Age Drives 3 Major Computing Trends
In the age of intelligence, Hu sees three major Computing trends that will shapes Huawei’s business strategy.
The first computing trend is that there will be a huge demand for incredible computing power.
“As computing moves from mainframes, to PCs, to mobile phones and now to wearables (in the age of IoT); Statistical computing, which is essentially a form of brute force computing, will eat up 80% of all available computing power in just five years time.”
This is the basis for Hu’s envisioning the next 10 years to be the ‘golden decade’ for Huawei’s opportunities in super computing.
The second trend is that – computing and intelligence will be increasingly ubiquitous .
“This means that computing power will not just be limited to brute-force computing (on the cloud), but present in everything at the edge such as on-device (ie. headphones and smartphones); to specialized-edge computing for things like gene sequencing.”
And the third computing trend is that in order to better serve the society and business community, there has to be effort to ensure that computing is managed cooperatively. In this sense, Hu said that the cloudshould only handle general-purpose (AI) model training, and providing background support for personalized on-device AI and specialized edge computing.
DA VINCI: Huawei’s Processor Architecture
Then there is Da Vinci, where Huawei has ambitious plans for its next-gen processor architecture to support all-scenario intelligence across device, edge, and cloud.
In its strong believe that computing power will be the bedrock for advanced AI, Huawei claims that its Da Vinci Architecture launched in 2018, will pave the way for future growth in the computing industry by supplying a steady resource of computing power “to be as readily available as oxygen” .
Da Vinci, says Hu, is the answer to form Huawei’s architectural processor foundation to address this anticipated hunger for super computing supply to the global market.
Focus is to Enable Partners, not Huawei per se, to Develop AI Apps
As though reading the thought bubbles of many in the audience, Hu addressed the unspoken question – Huawei will will steer clear from selling processors directly. Instead, its business strategy is to provide the computing power to customers in the form of cloud services, and to our partners in the form of technological components.
“ we won’t develop the (AI) applications ourselves, but we will provide tools and teams to the global partner ecosystem to develop specific AI applications that can solve problems in vertical markets – this includes providing our AI servers, accelerator cards, and modules for our partners, to integrate AI computing to enable application development and portability.”
Hu confirms that to aid the ecosystem’s effort for AI development, all of its software will be open-source – including server operating systems, databases, and AI development frameworks.
In 2015, the Huawei Developer Program was launched to help our partners develop better commercial software more easily. Hu shared that over the next five years, Huawei will invest another US$1.5 billion in its developer program.
“The aim is to expand the program to support five million developers and enable Huawei’s worldwide partners to develop the next generation of intelligent applications and solutions.”
For General Purpose Computing, KunPeng Calls.
A significant part of Huawei CONNECT 2019 is dedicated to lay the foundation for Huawei’s partner Kunpeng ecosystem. It builds on key technologies and products like accelerator cards, servers, operating systems, databases, compilers and other tools – all with the single key objective of enabling partners in the developer community to carry out AI application pilots, cultivate talent and develop standards.
So far, Huawei has Kunpeng innovation hubs in cities like Beijing, Shanghai, and Shenzhen; and Hu makes a welcome call on the more partners from all over the world to join KunPeng.
“Huawei is a company that is winning, and we take the long road as our mission is to bring best of technology to solve toughest problems. We believe such effort will expand the whole industry. We wish to see a thousand ships (ie partners) to sail together into the AI future, rather than one Huawei ship.”