Samsung Electronics is the latest mobile system-on-chip vendor to start using dedicated hardware for on-device machine learning. The Exynos 9 Series 9820, which will undoubtedly be used in the international versions of the Galaxy S10, has a neural processing engine (NPU) that is claimed to deliver up to seven times faster AI performance than the S9’s 9810. NPUs can be used to speed up tasks like image processing and AR.
Huawei was the first chip designer to announce an SoC with an NPU when it revealed the Kirin 970 last year, and Apple followed quickly with the A11 Bionic; both companies further focused on AI performance with their new Kirin 980 and A12 Bionic chips. Qualcomm, whose Snapdragon 845 chip is found in the US version of the Galaxy S9 and virtually every Android flagship around the world beyond Huawei’s, does not yet make dedicated AI-branded hardware, though it cited the 845’s combination of DSP, CPU, and GPU as delivering three times faster AI performance than the preceding 835.
20 percent faster single-core performance than the 9810
Elsewhere, Samsung is claiming the Exynos 9820’s fourth-generation custom CPU cores will provide a 20-percent boost to conventional single-core performance and 15-percent in multi-core or a 40-percent improvement in power efficiency. The GPU uses ARM’s new Mali G76 cores, first seen in the Kirin 980, for what is claimed to be 40 percent greater performance or 35 percent better power efficiency.
Samsung says the Exynos 9820 will be ready for mass production by the end of the year, which means it’ll be right on time for the rumored launch of the Galaxy S10 in late February. The question is now whether Qualcomm will follow the rest of the industry in focusing on AI for its 845 successor.