Course description
Accelerators such as GPUs have broken through in today’s HPC architectures. This course covers the more advanced topics in GPU programming with the CUDA language and the new OpenACC standard for directive-based accelerator programming. The course will focus on improving the performance and the scalability of GPU codes by analyzing them with performance tools, identifying bottlenecks, and making algorithm adjustments.Schedule
Tuesday, January 21
09:00 – 09:20

            Course introduction

            09:20 - 10:00

            CUDA recap I

            10:00 - 10:15

            Coffee/tea break

            10:15 – 10:45

            CUDA recap II

            10:45 – 11:15

            Introduction to CSC cluster

            11:15 – 12:00

            Cluster hands-on

            12:00 – 13:00

            Lunch break

            13:00 – 14:30

            Overview of CUDA on Kepler

            14:30 – 14:45

            Coffee/tea break

            14:45 – 15:30

            Tools and libraries

            15:30 – 16:00

            CUDA 6 overview

            16:00 – 16:30

            Debugging

            16:30 – 17:00

            Debugging hands-on

            18:00 -

            Dinner sponsored by NVIDIA (Restaurant Cantina, Otakaari 24, Espoo
    Wednesday, January 22
            09:00 – 09:45

            Profiling overview

            09:45 – 10:15

            Profiling hands-on

            10:15 – 10:30

            Coffee/tea break

            10:30 – 11:30

            Optimization 1

            11:30 – 12:00

            Optimization hands-on

            12:00 – 13:00

            Lunch break

            13:00 – 14:30

            Optimization 2

            14:30 – 14:45

            Coffee/tea break

            14:45 – 15:45

            Optimization hands-on

            15:45 – 16:15

            CUDA aware MPI

            16:15 – 17:00

            Optimization hands-on
    Thursday, January 23
            09:00 – 10:15

            Introduction to OpenACC

            10:15 – 10:30

            Coffee/tea break

            10:30 – 11:30

            Tuning OpenACC programs

            11:30 – 12:00

            Exercise session

            12:00 – 13:00

            Lunch break

            13:00 – 14:15

            Exercise session continues

            14:15 – 14:30

            Coffee/tea break

            14:30 – 15:30

            Advanced OpenACC topics

            15:30 – 16:15

            Exercise session

            16:15 – 16:30

            Course wrap-up
    Learning outcome

Understanding the GPU execution model and best practices in multi-GPU programming. Ability to parallelize and write scientific applications on GPUs.Pre-requisites
C/C++, Fortran, MPI, working knowledge in CUDAPrice: Free of chargeLecturers: Sami Ilvonen (CSC), Sami Saarinen (CSC), Peter Messmer (NVIDIA) and Axel Koehler (NVIDIA)

https://events.prace-ri.eu/event/216/