AMD Pulls the Rabbit Out of the Hat

Some time ago, AMD was touting a new revolutionary concept, called Torrenza, which included a second type of processor as a mathematical co-processor, capable of performing specialized calculations similar to a GPU. The plan revealed a two-processor platform, connected through the soon to come Hyper Transport 3 specifications, in which add-in co-processors could be used; these AMD Opteron based co-processors are also called "Accelerators".

Another big surprise came from AMD when they announced "Fusion", the project in which a CPU and a GPU merged as separate cores on the same processor die, thus bringing a different understanding to the term "integrated graphics". The connection between CPU and GPU is made through a crossbar and an integrated memory controller. Other than that, the CPU will have its own memory cache and the GPU will have separate access to its memory buffers, without interfering in each other's business.

AMD is bringing both of these technologies together, their goal being to create a platform with a minimum of two sockets, which will accept both accelerators and accelerated CPUs. This is also possible through the development of "stream computing", a technology which allows for parallel processing going past the Single Instruction Multiple Data (SIMD) instructions, and getting closer to Multiple Instruction Multiple Data. It uses general-purpose processors, which are able to execute floating point calculations, called stream processors.

The first processor made by using the stream computing concept was ATI Xenos, which powers the XBOX 360 game console. The new R600 is expected to use stream processing, and it will be AMD and ATIs first video card to be produced by them on this technology.

Basically, AMD wants to create an all-purpose computer capable of delivering a significant increase in processing power compared to present CPU architectures and also integrated graphics beyond what today's markets have to offer. In order to pull this off they would have to find a way to make all of these processors work seamlessly with each other. This is one of the reasons why AMD has made Hyper Transport an open standard, they need all the help they can get with this.