Murcia, España
In the first decade of the century, the Moore’s Law, which has led the microprocessor design in the last fifty years, was put into question by the scientific community. This was mainly due to the physical limitations of silicon-based architectures, which caused a change in the trend of designing processors, guided by parallelism. This transition has placed (massively) programming parallel as the only way to extract the maximum performance to new consumer platforms; being this essential to address today’s scientific challenges. Unfortunately, these challenges propose several issues whose computing needs are our of the scope for a single machine. Simulations, such as those discussed in this PhD Thesis, need to scale to large data centers; whose costs are only affordable for large institutions and governments. However, the current socio-economic situation requires an efficient use of resources. Tools such as cloud computing or volunteer computing offer an alternative to exploit computing resources in a flexible, fast, economical and environmentally friendly way. In my dissertation, we evaluate the current landscape of computation, previously described, using as a case study a high-impact problem for society as virtual screening. Virtual screening is a computational tool extensively used for drug discovery. The study has covered all processing levels, starting with an extensive analysis of the different commercially available alternatives at chip level, through their evaluation in a cluster environment, to scale to cloud computing and volunteer computing levels. This study concludes that GPUs are at the leading-edge of the development of scientific applications with massively parallel computing patterns and high computational demands, such as virtual screening.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados