Last year Nvidia launched it's new Fermi GPU architecture. Fermi was a shift from the GPU just being about graphics to taking on a bigger role as in general computing. Gamers were a bit concerned about so much die space now going to things that no longer directly improved gaming performance. But non-gaming industries have really taking a liking to the potential power of our graphics cards and Nvidia wants to sell products to them too.
Now it's AMD's turn. Graphics Core Next is a major architecture overall that tosses out graphics technologies used since the Radeon 9700 in order to make the GPU more capable at general computing.
GCN will include unified x86-64 memory addressing. This sounds like the discrete GPU will be able to access the system RAM via the PCI-E bus and CPU will be able to access the graphics card memory. Though of course there is a performance hit by using memory further away from the processor that needs it, so data will be copied to the memory that is closest to the processor doing the work.
Anandtech.com has a really thorough history lesson and analysis of what GCN is so far.
http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute
The presentation focused on the computing side of the GPU. But it's still a GPU and there will be improvements for gaming as well, however almost none of those details have been given yet. More to come later.
Anantech believes this technology may first come to high end GPUs in 2012 and would trickle down to lower end parts by 2013.
AMD is launching the Southern Islands GPU later this year, but it's mostly using the same architecture as the 6900 series, shrank down to a 28nm GPU and some tweaks.


