Conclusions and future trends

The Llano architecture is much more that a merge of a Propus class CPU and a Redwood class GPU.

There was implemented various energy saving techniques and hardware and software infrastructure capable of exploiting the possible synergies when two components are integrated on the same chip.

The performance/consumption ratio at platform level is increased both compared to previous AMD platforms, both with respect to the competitor platform, especially considering the applications which make use of GPU acceleration.

The chip real estate is a compromise in favor of the graphics in the AMD architecture, where it can boast more experience and know-how derived from the acquisition of ATI.

Besides, the future trend, thanks to the promotion of GPGPU, results in a greater use of graphics processor for non strictly recreational tasks.

The main reason is the vast computing power of the GPU, coupled with a quite acceptable consumption, which makes the performance/consumption particularly favorable.

This does not mean that the CPU will be abandoned, because there are tasks in which they are more efficient than GPU.

Driven integration between these two elements is desirable and the future in this direction: the Bulldozer architecture releases more emphatically the FPU from the rest of the core, setting the stage for possible future use of GPUs as a math coprocessor.

The missing piece is the unification of the address space between CPU and GPU, which is implemented in Llano at the driver level, but with subsequent architectures should be implemented at an even lower level.

Once these features were implemented, the term Fusion will take the essence of the AMD strategy.

Marco Comerci

CPU tests performed by Angelo Ciardiello