Current generation GPU architecture is characterized by an orientation toward the realization of a fixed set of functions defined by the DirectX API. Each new generation API, created with the participation of GPU developers, is an evolutionary improvement over its predecessor. It seems that AMD decided to depart from the accepted model, making the GPU architecture more open and universal.
Instead of models "tied up" to support DirectX, the company offers an increase the role of GPU to process data. And to a greater extent that is provided now via GPGPU mechanisms (using GPUs for general-purpose computing). The new approach has been presented at Fusion Developer Summit.
According to AMD, the future architecture GPU will fully support high level programming languages, including C and C + +. Integrated into the APU Fusion the corresponding blocks will be called "scalar coprocessors."
The architecture of the scalar coprocessor will combine the following elements: MIMD (multiple-instruction multiple-data: many streams of commands, many streams of data), SIMD (single-instruction multiple data, a stream of instructions, a lot of data streams) and SMT (simultaneous multithreading). Model VLIW (very long instruction word, a very long machine instruction) which had been used in several generations of AMD GPU's will no longer be there. No doubt that there will still be support of such API, as DirectX and OpenGL. Game developers can use familiar interfaces and C++ support for general applications. Of course, game developers will be able to go beyond what API models can do. Next generation Fusion GPU and CPU components will share a "truly common space of memory addresses."
0 comments:
Post a Comment