GPU Research Center - Università degli Studi di Perugia
Dipartimento di Matematica e Informatica

The University of Perugia has been selected by NVIDIA to be a GPU Research Center.

The research activity of the group operating at the Department of Mathematics and Computer Science of the University of Perugia, develops along different research lines, investigating the applications of NVIDIA hardware and CUDA to model and solve hard computational problems in computational logic, computational chemistry, systems simulation, etc. The design of new GPU-oriented algorithms is the most important contribution both to the realization of full-fledged systems and to the integration of GPU capabilities into existing tools for the mechanization of computational sciences.

The recent research activity is mainly focused on the following main themes:

Computational logic

On going research in computational logic focuses on the use of GPU computation capabilities in solving basic problems such as Boolean constraint satisfaction (SAT) and answer set computation (ASP). The relevance of SAT-solvers is well-known, it suffices to recall that establishing whether a SAT instance has a solution is an NP-complete problem. The same can be said for ASP. Efficient SAT/ASP-solvers are used as fundamental components of many tools for planning, problem solving, system configuration, automated reasoning, etc. In our group we implemented prototypical GPU-based tools for SAT/ASP solving. Their improvement through the development of new techniques, mainly based on graph algorithms are theme of current research.

Computational chemistry

We are active in the study and realization of CUDA-based computational programs for carrying out chemical/molecular simulations and study the properties of innovative systems and materials. This activity is developed in collaboration with the Chemistry, Biology and Bio-technology Department of our University.

Heterogeneous systems

Our group is also active in research on heterogeneous architectures and systems. We developed a simulation framework devoted to model and analyze scheduling problems and policies in hybrid/heterogeneous systems in order to improve their performance. New methodologies, exploiting cloud technologyes, are being defined for integrating GPU-enabled hosts in Grid environments. This enables jobs submitted to the Grid environment, to claim for GPU resources.