Description: Multicore processors are the current technological solution that keeps the CPU computational performance growing while satisfying the power and the cooling constrains imposed on high performance computing systems.
Moving with this trend HECToR now has 16 core AMD Interlagos processors. Two of these processors form a shared memory node.
Harnessing the high performance of the multicore CPU is not straightforward in many cases as the memory latency and bandwidth contention problems are exacerbated with the increase of the number of cores per CPU.
As solutions to these problems in this course we shall present the mixed mode OpenMP/MPI as a programming paradigm and the use of System V shared memory segments. The theoretical presentations will be supported by practical hands on sessions and attendees will also have the opportunity to discuss their own applications with HECToR CSE staff.
The specific issues regarding compiling and profiling for multicore processors is addressed in the Debugging, Profiling and Optimising course.
Aimed at: Programmers aiming their code at HECToR and similar multicore based system.
Prerequisites: Please note that this course does not teach a computer programming language. The target audience for this course is people who have done some parallel program development and would like to optimize their applications to run well on clusters of multicore processors. Therefore the course assumes a good working knowledge of Fortran or C and MPI as a pre-requisite. Students are also assumed be aware of OpenMP and be familiar with working in Linux, using text editors, and compliling scientific code in a command-line environment.
Duration: 1 day
After Course Attendees Will: Understand the implications of multi-core processors, and be able to exploit their performance.
Registration: To register for HECToR courses go to the booking form.