The third VSC Users Day was held at the "Paleis der Academiën", the seat of the "Royal Flemish Academy of Belgium for Science and the Arts", in the Hertogstraat 1, 1000 Brussels, on June 2, 2017.

Program

  • 9u50 : welcome
  • 10u00: dr. Achim Basermann, German Aerospace Center, High Performance Computing with Aeronautics and Space Applications [slides PDF - 4,9MB]
  • 11u00: coffee break
  • 11u30: workshop sessions – part 1
  • 12u45: lunch
  • 14u00: dr. Ehsan Moravveji, KU Leuven A Success Story on Tier-1: A Grid of Stellar Models [slides - PDF 11,9MB]
  • 14u30: ‘1-minute’ poster presentations
  • 15u00: workshop sessions – part 2
    • VSC for starters (by VSC personnel)
    • Profiler and Debugger (by VSC personnel)
    • Feedback from Tier-1 Evaluation Committee (dr. Walter Lioen, chairman) [slides - PDF 0.5MB]
  • 16u00: coffee and poster session
  • 17u00: drink

Abstracts of workshops

  • VSC for starters [slides PDF - 5.3MB]
    The workshop provides a smooth introduction to supercomputing for new users. Starting from common concepts in personal computing the similarities and differences with supercomputing are highlighted and some essential terminology is introduced. It is explained what users can expect from supercomputing and what not, as well as what is expected from them as users
  • Profiler and Debugger [slides PDF - 2,2MB]
    Both profiling and debugging play an important role in the software development process, and are not always appreciated. In this session we will introduce profiling and debugging tools, but the emphasis is on methodology. We will discuss how to detect common performance bottlenecks, and suggest some approaches to tackle them. For debugging, the most important point is avoiding bugs as much as possible.
  • Programming GPUs
    • Quasar, a high-level language and a development environment to reduce the complexity of heterogeneous programming of CPUs and GPUs, Prof dr. Bart Goosens, UGent [slides PDF - 2,1MB]
      In this workshop we present Quasar, a new programming framework that takes care of many common challenges for GPU programming, e.g., parallelization, memory management, load balancing and scheduling. Quasar consists of a high-level programming language with a similar abstraction level as Python or Matlab, making it well suited for rapid prototyping. We highlight some of the automatic parallelization strategies of Quasar and show how high-level code can efficiently be compiled to parallel code that takes advantage of the available CPU and GPU cores, while offering a computational performance that is on a par with a manual low-level C++/CUDA implementation. We explain how multi-GPU systems can be programmed from Quasar and we demonstrate some recent image processing and computer vision results obtained with Quasar.
    • GPU programming opportunities and challenges: nonlinear finite element analysis, dr. Vule Strbac, KU Leuven [slides PDF - 2,1MB]
      From a computational perspective, finite element analysis manifests substantial internal parallelism. Exposing and exploiting this parallelism using GPUs can yield significant speedups against CPU execution. The details of the mapping between a requested FE scenario and the hardware capabilities of the GPU device greatly affect this resulting speedup. Factors such as: (1) the types of materials present (elasticity), (2) the local memory pool and (3) fp32/fp64 computation impact GPU solution times differently than their CPU counterparts.
      We present results of both simple and complex FE analyses scenarios on a multitude of GPUs and show an objective estimation of general performance. In doing so, we detail the overall opportunities, challenges as well as the limitations of the GPU FE approach.

Poster sessions

An overview of the posters that were presented during the poster session is available here.