Here, emphasis is on powerful CPUs and strong communication between computer nodes. Everyone can pick up and deliver data via the parallel file system, Lustre, and also exchange data packets very quickly. Large calculations can thus be shared among several computers and take place many hundred times faster than if they were performed using a single computer.
DTU’s wind energy researchers have developed an algorithm specifically designed to solve the physical equations that describe wind flows.
Among other things, the calculations are used to predict how much energy a given wind turbine or wind farm will be able to produce in the course of a year. The computations can provide detailed data about speed, pressure and turbulence around wind turbine blades in all three spatial dimensions and time and link aerodynamic calculations with elastic structural ones. All this can be used to optimize the design and ultimately improve turbine service life and efficiency.
“Generally speaking, it is all about calculating the airflow around the wind turbine blade, and with ‘Jess’ we now often perform calculations using 100 million data points, which is a major improvement over ‘Gorm’—our previous supercomputer. Natural laws dictate that simply dividing the problem into several parts and calculating each one separately is not sufficient, as all the points are linked and impact one another. Currently, there are no HPC facilities sufficiently large to approximate physical reality 100 per cent, so we have to use turbulence models,” says senior researcher Dalibor Cavar, DTU Wind Energy.
‘Jess’ has 320 computer nodes, each with two CPUs and 10 cores. At the inauguration, ‘Jess’ was the among the largest supercomputers in Denmark, and at the same time one of the most eco-friendly. Energy accounts will also improve when the supercomputer moves to the new HPC centre, where the many watts are converted into heat via the water cooling system.