Parallel Computing on a CPU-GPU Cluster
Type: strategic reserve HES-SO
Target dates : 01.02.2010 – 31.01.2011
Total estimated cost : 190 kCHF
hepia/INIT Estimated cost : 110 kCHF
- HES-SO (hepia - leader, HEIG-VD)
- Department of computer science, University of Geneva
- Laboratory of Computational Engineering, EPFL
- APM Technologies Society (Geneva, Switzerland)
Aggregate optimization generally requires huge computing resources. The field of
bioinformatics, proteomics and genomics, illustrates this problem. Indeed, alignment or
protein folding algorithms are very greedy resources because on the one hand the volume of
data is huge and on the other hand these problems are part of difficult complexity classes.
The field of logistics, with the assignment of persons and equipment management, is also
representative and constitutes the CPUGPU project context. Indeed, the enterprise APM
Technologies develops software for airlines to generate flight programmes, to assign aircraft
flights and a crew on each flight.
The calculation time becomes quickly prohibitive even for companies with few airplanes.
However, for about two years, we have observed a craze for the use of graphic cards to
perform scientific computing.
The CUDA (Compute Unified Device Architecture) of NVIDIA encrypted in C language
allows the use of a graphic processor (GPU) as a CPU computation coprocessor.
The CPUGPU project aims to introduce an optimization module that runs on the GPU, to
improve the performance of the APM Technologies software. A subsequent parallel on a CPU
– GPU farm, will nevertheless increase performance.
Paul Albuquerque (firstname.lastname@example.org)