October 09, 2019 by Luke James
Chinese e-commerce conglomerate Alibaba unveiled its Hanguang 800 12nm AI chip, featuring a total of 17 billion transistors. At present, it is only for use in the company's own cloud servers.
Alibaba, the famous (or infamous!) Chinese e-commerce giant unveiled its first cloud AI accelerator chip at its Aspara Conference held in Hangzhou, China, on September 25. The chip, known as Hanguang 800, is a first for the company that has, until now, never launched its own silicon. However, the chip is not yet in production and Alibaba has no plans to sell it at present, opting to instead deploy it in its own cloud servers for customers to rent.
Alibaba CTO, Jeff Zhang, is reported to have said that, "The launch of Hanguang 800 is an important step in our pursuit of next-generation technologies, boosting computing capabilities that will drive both our current and emerging businesses while improving energy-efficiency,â€. The chip is reportedly designed to handle AI inference workloads.
It has been reported that Alibaba has put significant effort into ramping up its hardware efforts after the company's former CEO, Jack Ma, called for China to stop relying on U.S. imports. As you may well know, both the U.S. and China are presently locked in a bitter trade tariff war.
There are very few details surrounding the Hanguang 800's specs.
What is do know, however, is that the chip is based on a 12nm fabrication technology that crams 17 billion transistors together. Using ResNet-50, Alibaba has hinted at its performance by releasing a few benchmark figures.
These benchmark figures include the chip, when running ResNet-50, being able to process 78,563 images per second and operate at peak efficiency of 500 IPS per watt. Alibaba has said that this makes their own silicon 15 times faster than Nvidia's T4 GPU.
Alibaba CTO Jeff Zhang unveils the Hanguang 800 at the Aspara Computing Conference. Image Credit: Alibaba.
The Hanguang 800 chip is designed to carry out the type of calculations used by AI applications much faster than conventional silicon chips.
By developing the Hanguang 800 and using it exclusively to improve their own cloud network—at least for now—Alibaba hopes to provide its clients with access to advanced cloud computing and improve the search algorithms that are used on its e-commerce platform.
Alibaba now joins the likes of Google, Facebook, and Baidu, all non-traditional chipmakers, who are also designing their own AI chips in-house to improve the services that they offer. These companies are racing to build their chips as demand for computing power from AI technologies continues to grow.
By using RISC-V open-source architecture, it is much easier for such non-traditional chipmakers to develop their own silicon in-house. Previously, licensing fees would need to be paid to chip designers such as Arm.
Alibaba is a Chinese multinational conglomerate holding company specializing in e-commerce, retail, Internet, and technology. The company was founded in 1999 and provides business-to-consumer and business-to-business services via web portals, cloud computing, electronic payment processing, and others.
On the date of its IPO of $25 billion, the company's market value was US$231 billion, the highest in history.
