According to the data released by the two parties, Baidu's Kunlun artificial intelligence processor is mainly aimed at cloud computing, edge computing and other applications of artificial intelligence-like neural processor (XPU) architecture. The processor is expected to be built using Samsung's 14-nanometer process technology and its I-Cube TM packaging solution.
Baidu said that this processor will be able to provide 512GBps of memory bandwidth and achieve 260TOPS computing power at 150 watts of power. In addition, this new processor will also support Ernie's pre-trained model for natural language processing, making the overall inference speed 3 times faster than traditional GPU or FPGA acceleration models.
Baidu further pointed out that with the computing power and energy efficiency of this chip, Baidu can support a variety of functions including large-scale artificial intelligence budgets, such as search sorting, speech recognition, image processing, natural language processing, autonomous driving, and depth of PaddlePaddle. Learning platform.
Baidu said it is the first processor foundry cooperation with Samsung. Baidu will provide an artificial intelligence platform that maximizes the performance of artificial intelligence, and Samsung will expand its chip manufacturing business to high-performance computing (HPC) fields dedicated to cloud computing and edge computing.
In fact, in addition to Baidu itself actively developing dedicated processors, rival Google has also invested in this field for many years. Including the processors used by enterprises and cloud servers, Google has also made a considerable investment. Most of these processors are Tensor Processing Units (TPUs). Through the computing performance of these processors, Google can provide the industry's excellent computing capabilities to promote the development of machine learning and artificial intelligence training.(Responsible editor: admin)