CSC Autumn courses for LUMIΒΆ

The national supercomputer centre of Finland (CSC) in collaboration with PRACE is planning a series of courses in Autumn to teach the necessary development skills for the future users of the upcoming LUMI supercomputer. The initial courses are mainly oriented to programmers, as LUMI is first and foremost a machine to explore exascale technologies and hence new software. Nonetheless, there will also be a course program for non-developer users once LUMI becomes online starting early next year.

The training program for the CSC Autumn of HPC 2021 event proposes the following courses:

  • Parallel programming with MPI (September 2–3)

  • Advanced MPI and parallel I/O (September 16–17)

  • Hybrid CPU programming with OpenMP and MPI (October 4–5)

  • GPU programming with OpenMP (November)

  • GPU programming with HIP (December)

The first three of these courses focus on technologies for CPU programming and hence are not bound to any GPU architecture. The courses will cover open technologies, which are useful beyond LUMI. The only material that may be LUMI-specific is the chapter about compilation with the Cray Programming Environment used on LUMI.

The fourth course, GPU programming with OpenMP, covers the prime programming models for the AMD GPUs in LUMI. OpenMP provides shared memory parallelism, vectorization and offload to accelerators. This course should be useful for users of any GPU system as well, as OpenMP is an open standard well supported by the main manufacturers of GPUs.

The last course focuses on an AMD-specific CUDA-like technology, so-called HIP. This technology provides tools to port code written in CUDA, which is useful for users who want to continue doing CUDA-style programming yet have a migration path to AMD GPUs. Moreover, programs written in HIP can be easily compiled back on CUDA with native performance. This migration will be necessary for users of LUMI, since CUDA is a proprietary technology of NVidia and as such is not supported by its hardware.

On the VSC side, we expect to have two servers with current-generation AMD GPUs (the MI100) operational sometime in the fall. They will be integrated in the CalcUA and VSC infrastructure to support our users in their porting efforts to LUMI.

More information on these courses can be found on the CSC Autumn of HPC 2021 event page. Registration for the first three courses is already open.