Radeon Technologies Group - January 2016 - AMD Polaris Architecture

Graphics cards 1048 Page 1 of 2 Published by

teaser

Article Page 1 - Polaris Architecture

Radeon Technologies Group - Guru3D Technology update January 2016

The AMD Radeon Technologies Group is responsible for everything that is related to Radeon graphics cards and APUs. In this January 2016 update we can talk a little more about the new upcoming Radeon architecture which is dubbed, Polaris. 
 

Polaris

Polaris is the brightest star in the constellation Ursa Minor, and the 50th brightest star in the night sky. It is very close to the north celestial pole, Polaris is the North Star.
 

An embargo has lifted and as little information as it really is, today we talk about the architecture behind the Radeon GPU series that will launch in (likely) Q2 of 2016. There has been a lot of talk and gossip in regard to what fab node and process AMD will produce their new GPUs on. Will it be 14nm or 16nm ? To date that is still not 100% confirmed, it looks though that AMD will be using a combo of foundries for their 2016 chips. TSMC seems unlikely, yet Global Foundries is the first name that normally come to mind. Partly-confirmed now is that not the just GloFo, yet Samsung will be fabbing next years Arctic Islands family of GPUs alongside Global Foundries. Since it's Samsung this will be a 14nm FinFET fab, opposed to Nvidia who will be using 16nm for their Pascal architecture at TSMC.

During the latest RTG (Radeon Technologies Group) meeting Guru3D attended back in December 2015, the buzz talk was all about FinFET designs that allow for these smaller fabrication nodes which increases perf per watt and will allow for more transistors on a GPU or CPU package. The choice for Samsung Electronics and Global Foundries is not even that surprising as both share the very same IP for 14nm fabrication. This means the chips that are designed by AMD will be produced at both factories. When produced at both factories, AMD won’t have to worry about a problem regarding lack of supplies, anyone remember the Hawaii shortage once people found out that the 290/290X was hot stuff  for crunching bit-coins ? Spreading production also means spreading risk. Initially the first products that will roll of them wafers will be the GPU production of Greenland (Arctic Islands), AMD shortly thereafter will also volume fab its new CPU dubbed as Zen. This product will also be  based on a 14-nano LPP fab and again should be produced by Samsung Electronics and GF just like GPU.


Polaris Architecture

The codename for the new 14nm GCN revision 4 GPU architecture is Polaris. It seems we now have Arctic family series GPUs, with Polaris architecture, which is in fact 4th gen GCN architecture. A little confusing, sure. AMD opted Polaris as architecture name based on it's efficiency, Polaris, or the North star is one of the brightest stars we can see.
 

Untitled-2


We'll jump right onto that one as AMD claims gaming experiences at 61 percent less power than competing GPUs. The family of micro-architectures implementing the identically called instruction set "Graphics Core Next" has seen three iterations, the fourth will be used in Polaris. The new architecture features AMD’s 4th generation Graphics Core Next architecture, a next-generation display engine with support for HDMI 2.0a and DisplayPort 1.3, and next-generation multimedia features enabling 4K h.265 encoding and decoding. These chips will get updated compute engines, geometry processors, a different L2 cache and more. As to how far these all have changed remains to be seen as GCN up-to now only has seen small iteration improvements.

AMD - Polaris

Our guiding light is to power every pixel on every device efficiently. Stars are the most efficient photon generators of our universe. Their efficiency is the inspiration for every pixel we generate. 

Overall we expect differences in the instruction set to be minimal, micro-architectures often do not differentiate too much from one another. 
 

Untitled-8

Collection of different cores and engines that work together: multimedia – display – caches – memory controllers and power management.


Little information was shared and or is known about the new GCN iteration 4, however the Polaris GPU hardware is showing the architecture’s potential, in a recent event we attended AMD had a prototype on display. In a Star Wars: Battlefront head-to-head demo at 1080p, medium settings at 60 FPS, the Polaris GPU uses only 86 Watts, 61 percent less than the 140 Watts needed for the same experience on a GTX 950. Though that is comparing apples with oranges, at least from a perf/watt point of view, things look good.
 

Untitled-3


GCN 4th Generation "GCN 1.3/2.0" - GPUs of the Arctic Islands-family will be introduced in 2016 with AMD Radeon Rx 400 Series branded graphics cards. Any Arctic Islands-chip will be produced on a 14nm FinFET process. Currently the code-names that we know of to be used are Ellesmere Island, Baffin Island and Greenland.

An overview of key 4th gen GCN architecture relative to product changes are:

  • Primitive discard shader
  • Hardware schedulers
  • Instructions pre-fecter
  • Improvised shader efficiency
  • Memory compression
  • Next gen HDMI 2.0a and DP 1.3
  • Multimedia h.256 main 10 decode up to 4k / 4k h.265 at 60 fps encode.

On the next page we'll talk about the FinFET design, the underlying design of the transistors architecture that makes 14nm possible for a GPU.

Share this content
Twitter Facebook Reddit WhatsApp Email Print