Huawei Kirin 970 equipped with a dedicated unit for machine learning
This year, Huawei decided to dispense with the announcement of new devices at the IFA 2017 in Berlin. The main and in fact the only novelty was the flagship processor Kirin 970, which is expected in the upcoming Mate 10 and some models of the next year. The key difference of the Kirin 970 from its predecessors and solutions of competitors is the availability of a dedicated unit of the Neural Processing Unit, responsible for all kinds of operations related to neural networks, machine learning, computer vision and other things.
The performance of one Neural Processing Unit is 1.92 Tflops in half-precision calculations (FP16). For comparison, the 12-core graphic accelerator Mali-G72 MP12, which is part of the Kirin 970, has a performance of about 0.6 TFlops.
According to Huawei, Neural Processing Unit is 25 times more powerful and 50 times more energy efficient than traditional CPUs. Thanks to this, the Kirin 970 is able to recognize up to 2 005 images per minute, while other chipsets process about 97 images.
Specifications Kirin 970:
Technical process: 10-nm company TSMC;
Core: four energy-efficient Cortex-A53 cores with a frequency of 1.8 GHz and four high-performance cores Cortex-A73 with a frequency of 2.4 GHz;
graphics: Mali-G72 MP12, which is 25% more powerful than the one that was in Kirin 960;
artificial intelligence unit: Neural Processing Unit with a performance of 1.92 TFlops;
camera: dual image processor for capturing movements and better shooting quality in low light conditions;
video: decoding 4K @ 60, encoding 4K @ 30, support HDR10;
connection: support LTE Cat. 18 with a maximum data transfer rate of up to 1.2 Gbit / s;
Security: built-in Security Engine mechanism with support for TEE and inSE.
The first smartphone based on the Kirin 970 should be Huawei Mate 10, the presentation of which is scheduled for October 16.