by Anton Shilov
09/21/2012 | 03:49 PM
Advanced Micro Devices will continue to use names of islands as code-names for three forthcoming generations of its graphics processing units, it was revealed this week. Despite of the fact that the new chips' names will represent islands, the chips will be completely different and some of them promise to be truly revolutionary.
It is not a secret that AMD's next-gen family of graphics processing units (GPUs) is code-named Sea Islands. The year 2013 family of AMD Radeon HD graphics processing units is projected to be substantially more efficient than the Southern Islands when it comes to performance-per-watt and performance-per-transistor. However, there will be even more important architectural innovations, such as unified address space for CPU and GPU, GPU will be able to page system memory using CPU pointers, which will enable full memory coherency between the two.
Sea Islands will be made using now-proven 28nm process technology, hence we can expect a rapid ramp and more or less stable yields. What is noteworthy is that Sea Islands will hardly lower price on leading-edge GPUs significantly as Sea Islands chips are projected to be larger than their respective predecessors from the Southern Islands family.
In 2014, AMD will release rather revolutionary Volcanic Islands family (we are not sure whether it has a member code-named Eyjafjallajökull) of GPUs, which promises to become the pinnacle of heterogeneous - CPU+GPU - system architecture. Volcanic Islands GPUs, when accompanied by appropriate microprocessors and operating system, will support GPU compute context switch [GPU computes every single piece of application that it can] as well as GPU graphics pre-emption. Perhaps, accelerated processing units featuring Volcanic Islands GPUs with Excavator x86 cores will be able to do wonders by automatically using the most efficient execution unit (e.g., stream processors or x86 cores) for a particular task, which has potential to redefine the high-performance computing market (HPC), keeping in mind efficiency of stream processors in multi-threaded HPC apps.
It is known that AMD will be trying to maximally unify design methods of microprocessors and graphics chips. Potentially, this increases amount of potential foundry partners for each chip, which will not only let AMD to manufacture its VI chips using 20nm process technologies at different contract makers of semiconductors, but will enable the chip designer to quickly start manufacturing of accelerated processing units with Volcanic Islands GPUs inside.
The final "islands" family of GPUs will be Pirates Islands, reports VR-Zone web-site. Nothing particular is known about the family, which will become available in 2015. Given the naming and the timing, expect Caribbean code-names and chips made using 14nm process technology as well as further improvements when it comes to CPU+GPU collaboration.
AMD did not comment on the news-story.