This is now the next era – they simply called it Graphics Core Next. From a graphics standpoint, delivers cutting edge features and performance, while still being very flexible and scalable, allowing for all Southern Islands GPU parts to leverage the core. But as AMD has stated at AFDS, this first version of the GCN, delivers an amazing step up in terms of heterogeneous computing – both in terms of a new simpler and more powerful programming model, but also in terms of sheer efficiency and performance.
The GPU has 32 Compute Units and Dual Geometry Engines; there are also 8 Render Back-ends, up to 768KB read/write L2 cache which helps the compute part of the GCN. The memory runs on a 384-bit interface and the GPU itself is made on 28nm technology (4.3 billion transistors).
Thanks to the new architecture, we have much more compute performance than the previous generation:
Tesselation performance has also been increased and some in-house benchmarks show that:
The Partially Resident Textures technology that has been now introduced allows us to "treat" an ultra-large texture as if it were cached onto the GPU memory; now we can have textures the sizes of TB of information (uncompressed), sitting in a compressed format on the HDD and instead of loading all into the graphics memory, we can look first what the game engine is asking for, put only that into the graphics memory and render the scene(it will be available through an OpenGL extension):
Anisotropic filtering previous issues have now been fixed: