The workshop will consist of a full day of contributed and invited presentations and discussions on topics including, but not limited to:
– Current state of the art
- Today’s hardware
- Parallel numerical algorithms and libraries
- Emerging architectures that will enable exascale
- Effectively using millions of computational units
- Hybrid architectures
- Resilience and fault-tolerant algorithms
– Future directions
- Compiler directives, domain libraries, or code changes, or?
- Decomposition/coupling/multiphysics evolution?
- How can algorithm and library developers harness future architectures
Case studies, position papers proposing topics for discussion, and technical talks describing particular approaches are invited. The organizers seek participants from a broad spectrum of areas, including developers of scientific simulation codes and industrial end users. The presentations are expected to be 30 minutes in length, but we will entertain exceptions.
Simon McIntosh-Smith, head of the Microelectronics Group at University of Bristol, will talk about “Fault Tolerance Techniques for Sparse Matrix Methods”. Here you can find the detailed programme of the event (PDF).