World's Most Powerful AI Processor Meets the Market? Here Is How to Buy It

On September 18, HUAWEI CONNECT 2019 was opened in Shanghai.

This is one of Huawei's annual corporate-level flagship events. It has witnessed the release of many crucial strategies and products of Huawei. On HUAWEI CONNECT 2018, Huawei released its AI strategy and AI processors.

In fact, after Huawei released the world's most powerful AI processor, Ascend 910, on August 23, its business model has attracted high attention from the industry. In response to media questions, Eric Xu, Huawei's Deputy Chairman of the Board and Rotating Chairman, said that the Ascend 910-based training service could be launched to the market in China in September this year and globally in the first quarter of next year.

no.7 articlepreview image Huawei

Eric Xu, Huawei's Deputy Chairman of the Board and Rotating Chairman

The scarcity of computing power is one of the biggest pain points in the AI industry. Huawei's powerful computing chips can provide strong support for AI applications in various scenarios. But here comes the question: Can consumers and enterprises directly purchase the AI chips? In fact, all kinds of AI chips in the global market are basically provided to customers and partners in the form of modules and cards, servers, or cloud services. This is also how Huawei will do the business.

Executives from the Huawei Intelligent Computing Product Line say Huawei's Ascend series AI processors will be mounted on the Atlas series AI products and cloud services and launched to the market. Atlas is a series of AI computing products based on the Ascend AI processors, including modules, accelerator cards, edge stations, and servers. Atlas can be applied to terminal devices, edge computing, and cloud, meeting AI requirements in all scenarios and bringing pervasive intelligent computing. Currently, a large number of enterprises inside and outside China are using Huawei Atlas series modules and cards for AI product development.

The Atlas AI computing platform supports mainstream AI computing frameworks, such as Google's TensorFlow and Facebook's PyTorch/Caffe. It also supports Huawei's latest AI computing framework MindSpore. The MindSpore framework can implement algorithms as code, which greatly lowers the AI development threshold. In addition, the MindSpore framework has advantages in all scenarios. The deep collaboration between the MindSpore framework and Ascend 910 will fully unlock the computing potential of processors.

Currently, products similar to Ascend 910 are mainly provided by two companies, Google and NVIDIA. Google only provides cloud services but does not provide modules or cards.

In a media interview, Eric Xu also expressed the hope that global universities and research institutes can conduct research based on HUAWEI CLOUD services and Atlas series products with the Ascend 910 and 310 AI processors as the core. The biggest problem in AI research now is the scarcity of computing power. "Universities and colleges that have strong computing power will publish papers more quickly," according to Xu.

As shown by Huawei's official account on WeChat, China's largest social media platform, the Atlas AI computing platform series products have been launched on the Huawei Vmall. These products are integrated with solutions provided by industry partners and have been applied in scenarios such as facial recognition, license plate recognition, optical character recognition (OCR), and drone patrol.

AI computing is brute force computing.

The driving forces for AI development are algorithms, data, and computing power. It is difficult to quantify and track the innovation and development of algorithms, and it is also difficult to calculate the huge amount of data. However, computing power can be quantified.

Some in the industry even think that AI can rejuvenate Moore's Law because of the powerful computing capability. It can be said that computing power is the cornerstone of AI development, because sufficient computing power is instrumental to AI technology breakthroughs.

With the application of new technologies such as big data and AI, enterprises' requirements for computing power are soaring, and the traditional computing industry is in urgent need of transformation and upgrade. According to statistics, 20 ZB of new data is generated every year globally, and the AI computing capability requirement doubles every 3.5 months. This speed far outstrips the performance increase cycle defined by Moore's Law. Meanwhile, the rise of new technologies such as 5G, Internet of Things (IoT), and edge computing has also made computing requirements more diversified.

Although the algorithm efficiency is improved and the chip technology is highly developed, the computing power is still scarce. According to a study released by OpenAI, demands for computing power increase by an average of 10 times every year from 2012 to 2018. Especially in the 5G era, the data volume will increase explosively. The AI industry needs to cleanse, label, and train data on a larger scale. This means that the industry needs more time to perform computing, and insufficient computing power will impede AI development.

At the Ascend 910 release, commenting on the explosive growth of AI computing requirements, Eric Xu, Huawei's Rotating Chairman, made it clear that the biggest difficulty in AI research was the scarcity of computing power. In essence, AI computing is brute force computing.

Currently, key players in the AI server chip field include NVIDIA, Intel, Google, and Microsoft. More and more players are also entering the AI smartphone and IoT chip domains.

According to the statistics provided by Huawei, there are more than 1100 AI startups globally. According to Gartner's estimation, the market size of AI chips reached USD4.8 billion in 2017 and will reach USD14.6 billion in 2020. The market size of cloud applications is expected to reach USD10.5 billion.

All-scenario is Huawei's unique edge. Currently, Huawei is the only vendor in the world that can provide all-scenario AI solutions.

How strong is Huawei's computing power?

The Global Industry Vision (GIV) 2025 released by Huawei also predicts that the AI penetration rate in large enterprises will reach 86% by 2015. To address these challenges, Huawei has started to make full efforts in the computing field and is committed to providing customers with integrated solutions that are capable of powerful computing, cloud-edge collaboration, and all-scenario coverage by leveraging technologies such as chips, AI, and architecture innovation.

Ascend + Kunpeng sit at the heart of Huawei's powerful computing competitiveness.

In August this year, Huawei launched the most powerful AI processor, Ascend 910, and all-scenario AI computing framework, MindSpore, to complete its full-stack, all-scenario AI solution layout. With the support of Ascend series AI processors, the Huawei Atlas AI computing platform can be used to develop AI solutions for different application scenarios and industries to meet the computing requirements of different industries and scenarios.

Lab test results show that Ascend 910 has reached the computing power design specification, that is, delivering up to 256 TeraFLOPS at half precision (FP16), and 512 TeraOPS at integer precision (INT8). More importantly, the power consumption used to deliver such performance is only 310 W, which is much lower than the nominal 350 W of the design specification.

In the training of a typical ResNet50 network, Ascend 910 works with MindSpore. Compared with the existing mainstream single-card training with TensorFlow, Ascend 910 demonstrates a near 2x performance.

The emergence of Ascend 910 is a milestone. As the world's most powerful AI processor for a single chip, Ascend 910 is exactly made for AI applications. Based on the Ascend processors, Huawei has launched two series of products for AI applications and autonomous driving, respectively: Atlas and Mobile Data Center (MDC), to bring inclusive AI faster to reality.

Compared with the Ascend series, the Huawei Kunpeng 920 processor is mainly used for general computing. The TaiShan server based on the Kunpeng processor has the advantages of high performance and low consumption. In this way, enterprises can better process and calculate big data in real time, laying a foundation for the future intelligent era.

Based on the prediction of Huawei's full-stack, all-scenario AI strategy, the industry is now anticipating the Huawei server OS. Certainly such a release will infuse a new driving force into the computing industry.

So, how to buy such super powerful computing products?

When introducing the Ascend processor business model, Eric Xu pointed out that Huawei does not directly provide chips to third parties. Instead, Huawei provides chip-based hardware and cloud services. Huawei does not directly compete with chip vendors.

Ascend is sold to third parties in the form of AI accelerator cards, AI servers, and cloud services. Huawei will not sell standalone AI chip services.

In addition, Huawei hopes to cooperate with a large number of AI chip development enterprises to support various edge computing scenarios.

In April this year, Huawei announced the official launch of the Atlas AI computing platform series. Based on Huawei's Ascend series AI processors, the platform provides a variety of products, including modules, cards, edge stations, and appliances, to build an all-scenario AI infrastructure for the device-edge-cloud. It can be widely used in fields such as smart city, carrier, finance, Internet, and electric power. Integral to Huawei's full-stack, all-scenario AI solution, the Atlas AI computing platform unlocks supreme compute power to help customers embrace an AI-fueled future.

no.7 article image 2 Huawei

At the Ascend 910 release conference on August 23, Eric Xu said that Huawei had completed the layout of its full-stack, all-scenario AI solutions and delivered on the expectations and plans previously made. In Huawei's AI journey, this is a new milestone, and a new start as well.

Related:

Copyright © 2019 IDG Communications, Inc.