Child pages
  • Trends in HPC Environments
Skip to end of metadata
Go to start of metadata

Trends in HPC Environments

Paul C. Messina
TrackConference
DescriptionAbstract: The HPC computing ecosystem will be more encompassing than today’s.  As high-end computers approaches exaflops peak speeds in the next few years, the scientific and engineering computing ecosystem will need to enable not only more accurate and complex simulations. Modes of usage will evolve as well, requiring expansion of the software ecosystem typically found on supercomputers. Throughput workflows for ensembles, uncertainty quantification, and validation will constitute an increasing fraction of the workload in HPC facilities. Data-intensive projects will increasingly resort to in situ and streaming data analysis.  Archives of simulation and experimental data will be made available openly to worldwide communities, posing challenges in security and policies for longevity of the data.

Achieving good performance, especially strong scaling will be increasingly difficult because all systems will feature massive parallelism, less bytes of memory per flop, and a deeper memory hierarchy. Performance portability will be more important as well as more difficult to achieve, given the differences in the leading hardware architectures as well as the immaturity of programming models and much of the software stack.

Presentation documents