The next x86-compatible architecture of AMD after the current 28-nanometer architecture becomes Steamroller Excavator, also relies on 28-nm process. Excavator embodied in APU Carrizo during 2015. But in 2016 we can expect major changes on the microarchitectural front from AMD. Two years later, the company will present personally developed 64-bit ARM-architecture K12 and, as the cautious hints , new x86-compatible architecture, including a series of desktop processors FX.
Cheetah core will replace the core Puma +. Recently in 2015, will form the base of high-performance hybrid processor company. Consequently, hybrid processors core Cheetah appear at 2015-2016 years or, more likely, in 2016. What is interesting about the leak architecting AMD Cheetah was caused by the company's employees GlobalFoundries. This means that the new AMD architecture is adapted to the technical processes of this contract manufacturer of semiconductors, and orientation to 2016 may indicate plans to release APU Cheetah generation using 14-nm FinFET process.
The main feature of the architecture AMD Cheetah can become binding ARM-cores to its x86-compatible cores. In other words, every x86-compatible core get dedicated core architecture ARM. Obviously, 64-bit. Accordingly, the 2 -, 4 -, 6 - and 8-core APU generation Cheetah will also carry 2, 4, 6 and 8 ARM-cores. One of the objectives of ARM-cores may be to analyze the computational load on the core with x86-compatible architecture. Into the sphere of GPGPU-use calculations or heterogeneous computing. HSA heterogeneous architecture in the generation APU Cheetah will be called HSA +. Dedicated ARM core can in real time perform calculations switch GPGPU-computing flows on integrated graphics core and back. It is assumed that the technology will allow HSA + GPU to perform even those applications that do not use OpenCL. Another plus in the HSA + shared memory will be not only between the x86 CPU and GPU, but also between the ARM-cores.
As for graphics performance APU architecture Cheetah, it promises to be a generation GCN 2.0. Number of Stream pipelines will vary from model to model, reaching 1024 units in the most productive APU, and the frequency of senior video core can reach a value of 1000 MHz. Separately, AMD developers promise to take care of increasing the efficiency of ligaments discrete and integrated GPU. For the implementation of joint calculations and graphic Graphics Interchange will be provided a dedicated channel or bypass . It should be emphasized that the development of Cheetah is at a relatively early stage, so all changes are possible, including the abolition of the project.