The next-gen GPU architecture of AMD – Fusion evolutionby QArea Expert on June 21, 2011
Last week in Bellevue, Washington, AMD held the first Fusion Developer Summit (AFDS), here the company shared its vision towards the computing regarding the future of computing and the role graphics processors will play in it. In a few words, the company sees a future where it would be more facile for developers to create heterogeneous applications which efficaciously exploit not just the CPU but also the GPU’s heavy parallel processing capabilities.
The AMD company clarified that it presents moving concerning full system integration with the Fusion architecture. When running APUs are already integrated physically, with the GPU and CPU sharing a unified memory controller and the same silicon, AMD has an evolution strategy instead that will see the concept of APU grow even further.
That comprises of a quantity of evolutionary steps that are expected to endure through 2014, as for example the addition of support of C++ and other high level languages on the GPU to facilitate it for software developers to intensify its compute capabilities, empowering bi-directional power management among the GPU and CPU to boost performance or improve battery life, as well as accomplishing a unified memory address space and completely coherent memory parted by the GPU and CPU so that they operate evenly altogether.
As AMD’s Phil Rogers said future innovations are implied to make the various processor cores clearer for developers. They may then fluently tap into the power-efficient performance gigaflops that are available on the APU and even design faster, visually stunning apps on a great range of form factors.
It establishes a major overhaul of traditional GPU architecture of AMD and it will be applied throughout future generation products, together with discrete graphics cards and integrated GPUs with CPUs as component of AMD’s APU products. The concept is to build a GPU that operates well at both computing and graphical tasks, much like Nvidia is attempting to accomplish with Fermi and somewhat similar to what Intel bared in mind for the ill-fated Larrabee.
A detailed preview of as AMD is calling GCN – Graphics Core Next by AnandTech, that will be as GPUs architectural basis for the future GPUs of the company, detailing which way it operates and its features as well.