The Bifrost architecture, with improved support for high-end games on mobile and support for running machine learning applications, was introduced one year ago in May 2016 (see ARM’s Bifrost steps up graphics, bridges to machine learning and ARM takes VR/AR mobile with GPU core).
It is notable that in 2016 a billion chips shipped with Mali IP on-chip, up from 750 million units in 2015.
Now ARM has brought out an even higher performance Mali GPU based on multiple Bifrost optimizations including increased tile buffer memory, tiler scalability and L1 cache size.
The result is a 40 percent uplift in on-device performance compared with 2017 devices and 25 percent higher energy efficiency and 20 percent better area efficiency. While Mali-G71 was the first GPU to really support machine learning computation, its successor has been further optimized resulting in a 17 percent increase in machine learning efficiency, ARM claims.
Something called deferred shading is used based on pixel local storage and allows for the decoupling of geometry form lighting effects, which can otherwise be bandwidth intensive. The technique is common in high-fidelity gaming and bringing it to the mobile GPU helps those high-fidelity games be playing on the smartphone handset, which already has the basic screen resolution but needs to husband power resources.
Related links and articles: