We have released PyFR v1.15.0.
Improved performance and scaling of CUDA backend.
Improved performance across all backends via revised GiMMiK kernels.
Support for detecting load imbalances for multi-rank simulations.
Support for heterogeneous CPUs in the OpenMP backend.
Thanks again for your interest in the project.
3 October 2022 15:29
@mlaufer Are you okay to update the Spack packages?
Yes I am on it.
M1 Mac support through Spack is a bit troublesome but works with some workarounds.
3 October 2022 22:17
Also, feel free to submit a PR to the main PyFR repository explaining how Spack can be used in the quick install guide.
10 October 2022 07:41
This is really very awesome! I noticed that it got a new performance boost on the cuda backend of the program. I would like to know how this was achieved. How much of a performance improvement over the old version?
10 October 2022 12:17
The potential for a performance boost comes from the use of new GiMMiK kernels. If you will see a benefit is very situational, although it can be as much as 20%.