Last night, at IFA 17, Huawei Kirin 970 was finally official. It’s the world’s first built-in independent NPU (neural network unit) Artificial Intelligent SoC. The highlights are:
Kirin 970 for the first time uses TSMC 10nm process, integrating a 5.5 billion transistors (Snapdragon is 3.1 billion, Apple A10 is 3.3 billion), power consumption decreased by 20%. It’s based on four Cortex A73 running at 2.4GHz and four Cortex A53 cores at 1.8GHz. Moreover, the SoC is equipped with the 12-core Mali-G72 MP12 GPU. It supports:
In comparison with Kirin 960, Krin 970 CPU energy efficiency has increased by 20%, GPU performance by 20%, 50% reduction in power consumption (conversion down, energy efficiency increased by 140%).
At the same time, Kirin 970 joined the lower power consumption, ultra-high performance density NPU; image recognition speed goes up to about 2005 / min. It’s 5x times faster than iPhone 7 Plus. Huawei has invested more than 5,000 software engineers in optimizing NPU’s algorithms and experience.
More specifically, Huawei said that compared to the four Cortex-A73 core, in dealing with the same AI application tasks, the new heterogeneous computing architecture has about 50 times the efficiency and 25 times the performance advantage.
According to ARM design, Mali-G72 based on Bifrost architecture In comparison with the G71, G72 has 40% of the total performance increase (25% energy efficiency, 20% performance density), and this 970 from 8 nuclear to 12 nuclear.
Moreover, the Kirin 970 for the first time to support 4K HDR10, Dual ISP will drive three generations of Leica dual shot more powerfully.
Huawei Mate 10 will be the first phone to support Kirin 970 and will debut October 16. On that day, the first AI SoC true potential will be experienced around the globe.