With Huma unified memory architecture for AMD APU

AMD anticipates with the name Huma, heterogeneous unifom memory access, one of the main technical characteristics of the next generation of APUs belonging to the family Kabini

The integration in the latest processors such as Intel, AMD allows you to have a high computing power made available by the traditional x86 CPU core in combination with the components GPU stream processors with in-house.

For programmers to have the processing power so high, however, does not translate into simple approaches to exploit all this in a simple way and in parallel. CPU and GPU, in fact, do not share what is typically memory-space: this implies that in the presence of data that must be processed by the CPU as much as the GPU is necessary to perform various operations with copy space memory of the CPU to the GPU, and vice versa.

To overcome this limitation, and then allow to exploit the computing power of CPU and GPU, you need to be adopted unified memory space. Last year, AMD has anticipated that we will see this type of architectural novelty with family processors Kaveri, to debut during the second half of 2013. In recent days, AMD has confirmed this internal roadmap, indicating the name of Huma or heterogeneous unifom memory access.

With the solutions Kaveri work so we'll see a different approach than NUMA, non-uniform memory access, adopted so far by developed APU from AMD. With NUMA CPU and GPU share the same memory, but each of these components uses its own memory pool, and then the shared data must be copied from one memory block to another to be used.

With Huma GPU and CPU use a unified memory address space, sharing so much physical memory as virtual memory. The data consistency between the two different sources of processing is ensured without the need for synchronization operations, requests instead with the current NUMA approach: this should allow you to get both an increase in speed performance and a reduction in consumption.

We'll see how this evolution architectural, implemented entirely in hardware and therefore can operate with any operating system, will have a direct bearing on the speed performance of the systems. The first family of solutions Kaveri will debut during the second half of the year, in addition to implementing Huma also new Core x86 family Steamroller and GPU component derived Graphics Core Next architecture. All this will be made possible by the transition to 28-nanometer production technology, compared to the 32 nm used at the present time.