Baidu launches High-Performance AI Chip, Kunlun

PC: Baidu

Baidu has joined the league of Artificial Intelligence (AI) champions, Google, Nvidia, Intel along with many others with its announcement to launch of its cloud-to-edge Artificial Intelligence (AI) chip, Kunlun.

Baidu started developing field-programmable gate array (FPGA) chips in 2011 for deep learning use cases.

Back in 2011, Baidu focused on field-programmable gate array (FPGA) chips for deep learning use cases. It started using GPUs in its datacenter. The newly introduced Kunlun chip is 30 times faster than original field-programmable gate array (FPGA) chips. The exact specification of Kunlun chip are, 14nm Samsung engineering, 512 GB per second memory bandwidth and 260 tera-operations per second (TOPS) while consuming 100 Watts of power along with thousands of small cores, said the announcement.

The demand for computational power has increased drastically due to AI workloads. The main purpose of the Kunlun, an Artificial Intelligence chip, is large-scale AI workloads including voice recognition, search ranking, natural language processing, autonomous driving and large-scale recommendations. The company believes that Kunlun chip will allow for significant advancements in the open AI ecosystem and plans to further develop this chip.

Baidu creates “chip power” to meet the needs of various fields including intelligent vehicles, intelligent devices, voice recognition and image recognition, the company stated.

Kunlun chip will take advantage of Baidu's Artificial Intelligence (AI) ecosystem such as search ranking and deep learning frameworks like PaddlePaddle. The requirement to optimize the performance of these AI services prompted Baidu to build Kunlun Artificial Intelligence (AI) chip.

The announcement also referred to launch of Kunlun training chip “818-300” and inference chip “818-100”.  Data centers, public cloud, autonomous vehicles and edge scenarios can use Kunlun AI chip.