The recent financial analyst day for AMD was intended to be a turning point for the company’s fortunes, which have suffered from Bulldozer's ‘ahead-of-its-time’ performance and a lukewarm holiday season that has left most hardware vendors wanting.
One of the things it did bring was a lot of clarity to where Rory Read, AMD’s CEO, wants to take the company in particular with concerns to the CPU and GPU roadmaps.
Lisa Su, Senior Veep and General Manager for Global Business Units at AMD, presented a piece of slideware that shows what’s in store for discrete graphics in 2013, sort of.
2013 will see the debut of another nautically-themed family of graphics chips. “Sea Islands” is the name, a third generation DX11 part, second-generation Graphics Core Next architecture on 28nm that includes Heterogeneous System Architecture (HSA) features.
As AMD’s ticks and tocks are a bit out of sync – new architectures usually debut on mature processes, not on new ones, like GCN and its 28nm node – we’re inclined to say the soon-to-be 8000-series GPUs are based on a tweaked Graphics Core Next with added HSA features, rather than a revolutionary new architecture.
HSA, as AMD has put it, aims for a tighter integration of CPU and GPU functions within an APU, however Sea Islands is a discrete part, so we are keen on seeing where AMD is taking this. It also shows AMD will only release new parts next year and not late 2012 as some pundits wagered.
The first step will be to, apart from mobilising its software partners, leveraging whatever HSA feature the GPU has to AMD’s benefit, and this will likely be workload distribution between GPU and CPU plus memory management. From what we know of HSA, this is about all you can do right now on a discrete part.
Coming back to the name, “Sea Islands”, we won’t bet you the names of the chips, but you’ll excuse us if we say the Mediterranean is a better candidate than the Wikipedia definition of “a chain of tidal and barrier islands on the Atlantic Ocean coast of the United States”.
AMD will have to fend off Nvidia's Kepler until early 2013, it seems, but so far it seems in a good position to do so.