TSMC, which built both Apple’s and Huawei’s chips, went into volume production with 7-nm tech in April, and rival Samsung is moving toward commercial 7-nm production later this year or in early 2019. GlobalFoundries recently abandoned its attempts to develop a 7 nm process, reasoning that the multibillion-dollar investment would never pay for itself. And Intel announced delays in its move to its next manufacturing technology, which it calls a 10-nm node but which may be equivalent to others’ 7-nm technology.
Apple’s new A12 Bionic is made up of four CPU cores, six GPU cores, and an 8-core “neural engine” to handle machine learning tasks. According to Apple, the neural engine can perform 5 trillion operations per second—an eight-fold boost—and consumes one-tenth the energy of its previous incarnation. Of the GPU cores, two are designed for performance and are 15 percent faster than their predecessors. The other four are built for efficiency, with a 50 percent improvement on that metric. The system can decide which combination of the three types of cores will run a task most efficiently.
Huawei’s chip, the Kirin 980, was unveiled at the IFA 2018 in Berlin on 31 August. It packs 6.9 billion transistors onto a one-square-centimeter chip. The company says it’s the first chip to use processors based on Arm’s Cortex-A76, which is 75 percent more powerful and 58 percent more efficient compared to its predecessors the A73 and A75. It has 8 cores, two big, high-performance ones based on the A76, two middle-performance ones that are also A76s, and four smaller, high-efficiency cores based on a Cortex-A55 design. The system runs on a variation of Arm’s big.LITTLE architecture, in which immediate, intensive workloads are handled by the big processors while sustained background tasks are the job of the little ones.
Kirin 980’s GPU component is called the Mali-G76, and it offers a 46 percent performance boost and a 178 percent efficiency improvement from the previous generation. The chip also has a dual-core neural processing unit that more than doubles the number of images it can recognize to 4,500 images per minute.
Source: IEEE Spectrum