With advances in technology, FPU was integrated directly in the CPU, coprocessors, and (at least in the PC) , it ceased to exist as a separate device. But in 2006, AMD has revived the concept, which is found in a new device codenamed Torrenza.
From a technical point of view, Torrenza suggests a connection to the motherboard of a specialized calculations accelerator, which can increase system performance with certain types of tasks. The platform allows coprocessor connection not only through a special connector HTX, but also through the processor socket (if using a motherboard with two or more socket) or slot PCI Express.
After more than five years, Torrenza still not received a commercial embodiment. AMD leading expert on the software told that now the company do not see the point of release of graphic accelerators that are compatible with the socket. The concept Fusion, which involves the integration of computing with x86 cores stream processors Radeon in a single chip, according to AMD, is used on lshie prospects in high-performance computing.
The convergence of CPU and GPU, in fact, nullifying the effect of limitations of the standard layout, which is usually manifested in the insufficient capacity of the internal interfaces (PCI Express, for example) and high latency for data transmission. Realizing the potential of the ideal from a hardware perspective, APU falls squarely on the shoulders of programmers.
Curiously, the idea of Torrenza is reflected in the parallel processing accelerators from Intel, which offers the latest set of Xeon processors connectors.