Justification
This project aims to design and develop a general purpose compute cluster, based entirely on GPUs - originally intended for hardware acceleration. Though the project is supposed to build the system (in) the hardware, the main focus is to develop open source software solutions as well as libraries, tools and documentation that enable a broad range of use of such system, such as applications of image processing, video processing, simulation, mathematical computations, visualization of large volumes of data, especially applications that require real-time calculations. Due to recent advances and new architectures of GPUs, this theme is stirring up intense international interest in the scientific community. However the theme was unexplored and has massive potential for research. The development of a Brazilian solution for clusters of this type is strategic:
- A future convergence between the architectures of GPUs and CPUs with many cores. Therefore, strategies developed for distributed processing CPUs in the future may be appropriate solutions for many-core computers;
- The creation of a new concept of clusters of low-cost computing and processing power compatible with the current CPU-based clusters, whose cost can be more than 10 times the value of the solution proposed in this project will enable extensive research in optimization and high performance computing in research centers and laboratories with limited resources;
- The use of GPUs for general calculations is an extremely recent research topic, with a large scope of topics to be explored, as well as a large amount of research is still required.
In addition, the project aims at certain goals to be achieved:
- Expand the scope of applications using GPUs for processing generic;
- Consolidation of cooperation and exchanges;
- Publications in scientific journals of the events and achievements.