Graphcore joins Baidu PaddlePaddle hardware ecosystem to achieve breakthrough innovation in machine intelligence

At today's Wave Summit 2020 Deep Learning Developer Summit, Baidu officially announced Graphcore as one of the partners in the Baidu PaddlePaddle hardware ecosystem co-construction plan, and jointly signed a proposal to help AI innovative applications land in various scenarios and promote the establishment of unified industry standards. Through Baidu PaddlePaddle, more Chinese developers will be able to use Graphcore IPU, a new processor architecture, to achieve breakthrough innovations in machine intelligence and significantly accelerate AI models.

Wu Tian, ​​Vice President of Baidu Group and Deputy Director of the National Engineering Laboratory for Deep Learning Technology and Applications, released the PaddlePaddle hardware ecosystem

Baidu PaddlePaddle is China's first open-source, fully functional industrial-grade deep learning platform. It currently has more than 1.9 million developers, serves 84,000 companies, and has created more than 230,000 models, making it a leader in China's deep learning platforms. The establishment of Baidu PaddlePaddle's hardware ecosystem aims to accelerate the popularization and promotion of deep learning and artificial intelligence applications in the Chinese market. There are 13 initial members of Baidu PaddlePaddle's hardware ecosystem, covering different hardware manufacturers on the cloud and device side. Among them, Graphcore, as an important partner of Baidu PaddlePaddle in cloud training and reasoning, will help developers achieve significant acceleration of AI innovation models through IPU technology in the cloud and data center, thereby gaining an advantage in the market.

 "The integration of intelligent hardware platforms and deep learning frameworks has become the key to building a world-leading AI application and promotion ecosystem. Through the Baidu PaddlePaddle hardware ecosystem, we hope to work with partners to accelerate the integration of hardware and software. Graphcore is an important partner of Baidu PaddlePaddle in cloud training and reasoning. Through the PaddlePaddle hardware ecosystem, more developers using PaddlePaddle can use Graphcore's IPU technology for machine learning innovations, shorten training time, and improve development efficiency." said a relevant person in charge of Baidu PaddlePaddle.

Mr. Lu Tao, Vice President of Sales and General Manager of China at Graphcore, said: "Baidu PaddlePaddle is China's leading deep learning platform. We are very pleased to work with Baidu PaddlePaddle to build a hardware ecosystem and accelerate the popularization of various AI applications. Graphcore designed the intelligent processor IPU from scratch, which can be used for both cloud training and inference, aiming to help AI innovators break through the bottlenecks brought by traditional hardware. Graphcore's IPU hardware and solutions have been mass-produced and applied in different AI scenarios. Through Baidu PaddlePaddle, more developers can use Graphcore IPU, a new processor architecture, to greatly accelerate AI models and make breakthroughs in the next wave of machine intelligence. In the future, we will continue to deepen R&D cooperation with Baidu PaddlePaddle to accelerate the adaptation and implementation of algorithm models such as machine vision and natural language processing."

Tao Lu, Vice President of Sales and General Manager of China, Graphcore

Faced with the new global normal, the computing demand in the cloud has exploded from computing volume to computing power. CPUs and GPUs were never designed to meet the computing needs of machine learning, so cutting-edge innovative AI algorithm models are often constrained by hardware and forced to compromise. IPU is completely different from today's CPU and GPU processors. It is designed from scratch by Graphcore and is specifically suitable for computing-intensive machine learning and deep learning tasks. It is a highly flexible and easy-to-use parallel processor that can achieve state-of-the-art performance on machine intelligence models currently used for training and reasoning. IPU and the product-ready Poplar software stack provide developers with powerful, efficient, scalable and high-performance solutions to help realize AI innovation. By accelerating more complex models and developing new technologies, customers can solve the most difficult AI workloads. At present, IPU products have entered mass production and delivered to global customers, and are used in AI scenarios such as finance, medical care, telecommunications, and the Internet.

Graphcore is committed to developing and deploying machine learning applications and models more quickly and easily from existing frameworks, supporting the next generation of machine learning by providing users with the ability to program directly at the hardware level. Therefore, Graphcore can help PaddlePaddle further lower the development threshold and improve development efficiency. PaddlePaddle users are mostly senior AI practitioners who have high requirements for computing power, and many of them deal with highly computationally intensive tasks such as natural language processing, computer vision, and video analysis. They can create their own machine intelligence models on the IPU, program directly at the hardware level without sacrificing ease of use, and achieve rapid operation and iteration of innovative models.

In the future, as a member of Baidu PaddlePaddle’s hardware ecosystem, Graphcore will work with Baidu PaddlePaddle to:

• Accelerate the integration of software and hardware: to meet the needs of cutting-edge research and industrial applications of typical AI application scenarios such as vision, natural language processing, and voice, and to accelerate the integration of deep learning frameworks in training, prediction and other functions with chips, complete machines and other intelligent hardware platform manufacturers.

•  Establish unified industry standards: Promote the establishment of unified industry standards in the fields of deep learning frameworks, software and hardware adaptation interfaces, and whole-machine system integration, and promote large-scale application in the industry.

• Work together to promote the popularization of results: Build and improve the operating mechanism of the deep learning and artificial intelligence industry application ecosystem, and work together to promote the accelerated popularization of deep learning and artificial intelligence application results in the Chinese market through cooperative competitions, industry-university-research integration, and other forms. 

About PaddlePaddle:

Baidu PaddlePaddle is a domestically independently developed, open source, and most comprehensive industrial-grade deep learning platform that integrates a deep learning core framework, basic model library, end-to-end development kit, tool components, and service platform. PaddlePaddle originated from industrial practice and is committed to in-depth integration with the industry. It provides leading deep learning and machine learning task development, training, and deployment capabilities to accelerate the process of enterprises from algorithm development to industrial implementation. Currently, PaddlePaddle has been widely used in industry, agriculture, service industries, etc. There are more than 1.9 million developers developing on the Baidu Brain Open Platform and PaddlePaddle platform, and together with partners, it helps more and more industries to empower AI.

About Graphcore®

Graphcore IPU (Intelligence Processing Unit) hardware and Poplar® software help innovators create the next generation of machine intelligence solutions. IPU is the first processor designed specifically for machine intelligence. Graphcore has raised over $450 million in funding from leading financial and strategic investors and is headquartered in Bristol, UK.