Given the notable growth of peak performance of current HPC systems compared to the bandwidth with which data can be transferred to the compute units, CFD codes are getting more and more memory-bound. That is, the bottleneck is not how fast we can compute stuff, but how fast we can transfer the data. In order to deal with this, we can run multiple simulations simultaneouslly in the same device, so that we become more efficient in handling the data.
This was the main idea behind the work presented in the 9th European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS), which had as a title “Performance analysis of parallel-in-time techniques in modern supercomputers”, related to the research done in CTTC – UPC | Universitat Politècnica de Catalunya – Centre Tecnològic de Transferència de Calor.
I would like to acknowledge my colleagues F.Xavier Trias, Àdel Alsalti, Guillem Colomer and Assensi Oliva for their contribution in this work, as well as the organizers of the minisymposium in which the work was presented.
Scroll to Top