Objectives

In order the possibility developed and available for interconnectivity of GPUs, the project's main goal is to develop a solution for all clusters based on such hardware. Solutions of this type already exist in the market, but not with open architecture..

The research aims to create a solution that is compatible with typical cluster based on CPUs, but at a reduced cost. In addition, we intend to explore certain kinds of calculations and simulations, especially focused on applications requiring answers in real time, where the solution is more efficient using GPUs those CPUs.

In addition to the physical construction of a cluster, the project aims to develop tools and libraries, with free license for the development of applications on these platforms. Despite the trend in GPUs have become general purpose processors, still have a different programming paradigm for the development of CPUs

The libraries and tools intended to be developed to encapsulate the development process for the application programmer, so that applications can only be made using paradigms and traditional development environments. The system will perform an internal redirection mapping and code snippets for the cluster of GPUs, allowing the programmer only reason in relation to traditional parallel algorithms, the same that would be used in case of using a traditional cluster, based on CPUs.

In the case of CPUs cluster, due to the generic nature that they present in their internal architecture, any processing can be performed by any node. In the case of graphics hardware architectures, it is not trivial, since there are nodes that are optimized for specific calculations.

Thus, this project also has the following objectives::