China’s Google, Baidu Inc. made public its first self developed AI chip named ‘Kunlun 818-300 model chip’ at the annual ‘Baidu Create AI developer conference‘. The company joins firms like Google, NVIDIA, Intel and others in firms specializing in machine learning.
Kunlun is optimized for various AI tasks, including voice recognition, search ranking, natural language processing, image recognition and autonomous driving. Owing to its great processing power Kunlun can provide both cloud services and edge functionality. Hence, it can be used in powering Baidu’s self-driving bus, Apolong. Baidu developed its self driving bus in partnership with King Long and is powered by Apollo. It aims to put Apolong into commercial operation in Xiongan, Beijing, Wuhan, Shenzhen and Pingtan of Fujian province.
Baidu in 2011 first started working on its customized AI processors using FPGAs and began using GPUs in data centers for deep learning. In his talk, Robin Li Yanhong, founder of Baidu bragged that Kunlun is “by far the most powerful AI chip in China’s industry. The chipset was born from a massive AI algorithm, which is also based on Baidu’s eight years of research on CPU [central processing unit], GPU [graphics processing unit] and FPGA [field-programmable gate array] technology” he said. The new design is thirty times faster than the original FPGA-based processor. Kunlun is able to achieve 260 tera-operations per second (TOPS) and 512GB/second memory bandwidth, according to VentureBeat. Other key specifications include: 14nm fabrication process engineering, 512 GB/second memory bandwidth and 100 watts of power consumption.
This is a substantial innovation and with increasing demand of high-performance creates a need for cost-effective solutions which can fulfill the demands of machine learning application.
“With the rapid emergence of AI applications, dramatically increasing requirements are being imposed on computational power,” the company said “Traditional chips limit how much computing power is available and thus how far we can accelerate AI technologies. Baidu developed this chip, specifically designed for large-scale AI workloads, as an answer to this demand.”
As part of this, Baidu plans to continue its effort on developing this chip, developing it progressively to enable the expansion of an open AI ecosystem. Untill it reaches the so called ‘chip power’ it will meet the needs of various fields and then transform its AI research results into marketable products and services. “Baidu needs to focus on not only the performance of its chipset but also a build-up of the AI system, which means how many companies will use its chipsets” Xiang said. Still, it has advantages in its algorithm structure” he added. Connecting chipset package with different algorithms will definitely make the company more competitive.
Baidu Create AI developer conference also witnessed announcement of Baidu Brain 3.0. Haifeng Wang, Baidu Senior Vice President and Head of Baidu AI Group, explained: “The core [of Baidu Brain 3.0] is its multimodal deep semantic understanding. This means that for modal information such as images, voice, and video, machines can not only hear and see, but also understand the meaning behind it.” Baidu Brain 3.0 would add a great advantage to AI services provided by the firm.
The release of these new AI powered chips sends a strong message around the globe. The message being unlike earlier, Chinese internet companies are taking semiconductors seriously and are expanding their business, from Internet to AI chips to integrated smart products.