Energy Efficient Applications
As Moore’s law continues to hold and computing systems double their performance roughly every 18 months, applications need to cope with ever increasing levels of parallelism and efficiently adapt to increasing memory hierarchies (CPU registers, L1-L3 cache, main memory, background storage). An increasing number of modern high performance computing systems also use compute accelerators such as graphic processing units or Intel’s Many Core architecture Xeon Phi to achieve even higher peak performance. This increase in complexity of today’s HPC architectures results in an increased porting effort to run applications efficiently on the compute systems.
In close collaboration with the ‘Munich Centre of Advanced Computing’ (http://www.mac.tum.de/wiki/index.php/Main_Page) LRZ engineers are developing tools for automatic profiling of application performance and efficiency. Selected applications from the domains of astrophysics, geophysics, and life sciences are ported to new HPC system architectures and further optimized in so called ‘LRZ application labs’