AMD Crystal Ball

Recently we witnessed an alleged leak for the first member of the Sea Islands Family, which is scheduled to appear in the first quarter of 2013. We dug deeper and discovered codenames for GPUs coming in the next couple of years.

The first leaked specifications typically represent a ballpark figure or a performance target. Sometimes, those numbers are beaten by a significant margin, sometimes you have to go back to the drawing board and redesign PCB, firmware, timings and optimize drivers to address the performance shortfall.

In terms of graphics processors, we've enjoyed more or less stable environment with pleasant (NVIDIA Kepler/GK104 beat internal performance projections by 20%) and unpleasant (the original Fermi/GF100 chips came with broken fabric and did not work) surprises. When we take a look at AMD's camp, the company set on its path in 2008-2009 timeframe, when they decided on a long sequence of Island-named GPU architectures. The complete cadence is scheduled to complete in 2016, when the GPU should have equal abilities of communication as the CPU and directly address CPU routines which can be accelerated.

At least, that's what our sources are saying – "It doesn't matter if its ARM, x86 or even MIPS. GPU will extend processing (capabilities) of every CPU from tablet you have in your hands to 42U racks in datacenters". If you take a look at the official AMD slide from the Financial Analyst Day 2012, you can see where the company is heading.

 

2013: Sea Islands

Recently, website Videocardz leaked out first details from the 2013 GPU "Sea Islands" architectural family. As a successor to Northern (Hemisphere) Islands from 2011 and Southern (Hemisphere) Islands from the current year, "Sea Islands" (SI) will feature codenames from seas, rather than oceans. Thus, it is of no surprise that Radeon HD 8800 family carries codename "Oland", which is Sweden's second largest island.

Oland should come as a 3.4 billion transistor chip, carrying 20% more transistors than its predecessor Pitcairn (HD 7800). The number of GCN units is not known at this time, but the performance is targeting GeForce GTX 680. Manufactured in 28nm process, AMD engineers worked hard on optimizing transistor libraries at TSMC in order to increase transistor efficiency, paying the price with an increased die size. This is the same path NVIDIA followed for years, sacrificing smaller die size for improved yields.

However, performance will take second place to the amount of architectural changes Sea Islands bring. First and foremost, Sea Islands will bring Unified Address Space between the GPU and the CPU. Furthermore, GPU will be able to page system memory using CPU pointers, bringing full memory coherency between the two. This is something even Intel doesn't have planned for its Xeon Phi (Larrabee / Knights' Corner / Knights' Ferry).

Sea Islands will spend better part of the year without the real competitor, as NVIDIA will only offer refreshed Kepler until Maxwell comes along. Also worth mentioning is that SI is the architecture chosen to be expanded into high-performance consoles, thus we should see quite interesting announcements regarding to vast compute and graphics capabilities carrying the next generation of console games. Bear in mind that engines such as Unreal Engine 4 do not work on contemporary consoles, while PC hardware can easily churn out those pixels in ultra-high details.

 

Bring on 2014: Volcanic Islands

After Sea Islands reaches the end of the line, AMD will introduce perhaps the most important GPU family of them all – Volcanic Islands. Volcanic Islands (VI) will go head to head against NVIDIA Maxwell and second-gen Xeon Phi architecture.

Manufactured at 20nm Gate-Last process, this will be the first GPU family which AMD should be able to manufacture in Common Platform Alliance as well as its long-standing foundry partner, TSMC. Thus, AMD will have the choice between TSMC GigaFab Hsinchu/Taichung, IBM East Fishkill, GlobalFoundries in New York and Dresden or Samsung in Austin. The manufacturing flexibility will be of paramount importance, for Volcanic Islands GPU architecture will represent the pinnacle of system integration between the CPU and GPU. In terms of Fusion/FirePro APUs, which will replace the current FX/Opteron CPUs, Volcanic Islands bring silicon-level system integration. APU will extend to the Discrete GPU and treat it as one, as well as VI silicon treating the CPU as an integral part.

 

2015: Pirates of… Graphics?

Perhaps one of most interesting codenames is coming down the pipe in 2015: Pirates Islands. Yes, you've read it correctly. Bringing a tribute to legendary pirates such as the Blackbeard, Captain Hook or well, Captain Jack Sparrow, AMD's imaginative engineers are targeting the 20nm process with 14nm APUs in mind. The real work on this GPU architecture only started recently (GPUs are being designed 2-4 years in advance) and guessing what lies on cards is a bit too unpredictable. We know feature set, but performance… your guess is as good as ours.

 

When it comes to codenames for mobile parts, you can expect that near future brings departure from this planet, but that's all we're going to disclose… for now.