AMD unveils 14nm GCN 4.0 'Polaris' GPU architecture

January 5, 2016 | 11:37

Tags: #gcn #gpu #graphics-core-next #performance-per-watt #polaris #radeon

Companies: #amd

AMD has revealed details of its upcoming Polaris GPU architecture, promising double the performance per watt thanks to a redesign and a new 14nm FinFET process node, as it begins sampling the parts to its hardware partners.

'Our new Polaris architecture showcases significant advances in performance, power efficiency and features,' claimed Lisa Su, president and chief executive officer of the financially ailing AMD at the first public demonstration of the Polaris architecture, the fourth generation of the company's Graphics Core Next (GCN) family. '2016 will be a very exciting year for Radeon fans driven by our Polaris architecture, Radeon Software Crimson Edition and a host of other innovations in the pipeline from our Radeon Technologies Group.'

Headline features for the Polaris architecture, as unveiled by AMD during the demonstration, include support for HDMI 2.0a and DisplayPort 1.3 output as well as 4K-resolution h.265 encode and decode in hardware. The big win, though, is in performance: a shift to a 14nm process node using FinFET three-dimensional transistor technology has led to a considerable boost in efficiency, with the company claiming that a mid-range Polaris GPU will be able to produce a smooth 60 frames per second in Full HD for an 86W power draw from the wall when installed in an Intel Core i7 gaming rig, compared to 140W for a system running Nvidia's GeForce GTX 950.

This gain in performance per watt isn't just down to the new process node, though: the company claims to have made major improvements to the Graphics Core Next architecture for Polaris, including new memory compression capabilities, improved shader efficiency, a primitive discard accelerator, a hardware scheduler, and instruction pre-fetch. All told, of the twelve components to appear in the Polaris block diagram, seven are claimed to be in some way 'new' - with only the global data share hardware, compute engines, scheduler, render backend and rasteriser largely untouched compared to GCN 3.

Full specifications and performance figures have not yet been provided by AMD, which has stated that it expects to see the first Polaris-based desktop and laptop hardware on the market in around six months. Meanwhile, the company's presentation on the new architecture is reproduced below.

Discuss this in the forums
YouTube logo
MSI MPG Velox 100R Chassis Review

October 14 2021 | 15:04