Baidu Reveals Kunlun AI Silicon Chips For Edge And Cloud Computing

AI needs a lot of computing power. And with the growth of cloud computing, the demands for even more sophisticated computers continue.

Baidu is a Chinese multinational technology company, and it has revealed China's first cloud-to-edge AI chip for training and inference. With the name 'Kunlun', the chip is made up of thousands of small cores. It sports a Samsung 14nm engineering capable of achieving 260 tera-operations per second (TOPS) and 512 GB/second memory bandwidth.

All while consuming 100W of power.

Kunlun is built to acomodate high performance requirements of a wide variety of AI scenarios, as well cloud and edge scenarios, including data centers, public clouds and autonomous vehicles. The announcement includes training chip "818-300"and inference chip "818-100".

Baidu's years of experience in optimizing the performance of AI services and frameworks has given the company the expertise to build this kind of AI chip.

Baidu Kunlun AI chip
Kunlun AI chip, debuted in at the Baidu Create AI developer conference in Beijing

Baidu said that:

"With the rapid emergence of AI applications, dramatically increasing requirements are being imposed on computational power. 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."

The Chinese company believes that Kunlun will allow significant advancements in the open AI ecosystem.

In 2011, Baidu started working with field-programmable gate array (FPGA) chips especially designed for deep learning, and began using GPUs in its data centers. And Kunlun here, has a computational capability which is nearly 30 times faster than FPGA.

And with Kunlun, Baidu in joining the likes of Google, Nvidia, Intel, and other tech companies making processors especially for AI.

As the first of its kind in China, Baidu plans to continue the development of the chip, progressing it to enable the expansion of the AI ecosystem. As part of this effort, Baidu is also planning to create "chip power" to meet the needs of various fields including intelligent vehicles, intelligent devices, voice recognition and image recognition.

Baidu also announced an upgrade to its suite of AI services, and released Baidu Brain 3.0

With it, the platform can supply simple drag-and-drop training for AI models. It also boasts 110 AI services ranging from computer vision to natural language processing to facial recognition software.

The announcements were made at the second annual AI developer conference, Baidu Create 2018, in Beijing.

Published: 
04/07/2018