Rempke, Arne und Krzikalla, Olaf und Mohnke, Jasmin und Wendler, Johannes und Wagner, Michael und Gerhold, Thomas (2023) Spliss: A Sparse Linear System Solver for Transparent Integration of Emerging HPC Technologies into CFD Solvers. Computational and Data Science Seminar, 2023-01-24, Jülich, Deutschland.
PDF
3MB |
Kurzfassung
We introduce Spliss, a block sparse linear solver library, which is designed to meet current and upcoming challenges in the CFD development. The C++ template interface of Spliss supports a wide range of linear operators typically used in CFD applications. This includes sparse block matrices of fixed or variable block sizes, different scalar types and matrix-free operators. We explain the technical design of Spliss, by which we can achieve computational efficiency and parallel scalability on contemporary and emerging HPC platforms. The presented performance evaluation is conducted on the DLR cluster CARO and on Juwels Booster. It shows how we have passed the tipping point, beyond which GPUs perform strictly better, by integrating GPUDirect and CUDA-aware MPI communication.
elib-URL des Eintrags: | https://elib.dlr.de/193786/ | ||||||||||||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Dokumentart: | Konferenzbeitrag (Vortrag) | ||||||||||||||||||||||||||||
Titel: | Spliss: A Sparse Linear System Solver for Transparent Integration of Emerging HPC Technologies into CFD Solvers | ||||||||||||||||||||||||||||
Autoren: |
| ||||||||||||||||||||||||||||
Datum: | 24 Januar 2023 | ||||||||||||||||||||||||||||
Referierte Publikation: | Nein | ||||||||||||||||||||||||||||
Open Access: | Ja | ||||||||||||||||||||||||||||
Gold Open Access: | Nein | ||||||||||||||||||||||||||||
In SCOPUS: | Nein | ||||||||||||||||||||||||||||
In ISI Web of Science: | Nein | ||||||||||||||||||||||||||||
Status: | veröffentlicht | ||||||||||||||||||||||||||||
Stichwörter: | Sparse Linear Algebra, Iterative Solvers, Performance, Spliss, HPC, CFD, CODA, CUDA, MPI | ||||||||||||||||||||||||||||
Veranstaltungstitel: | Computational and Data Science Seminar | ||||||||||||||||||||||||||||
Veranstaltungsort: | Jülich, Deutschland | ||||||||||||||||||||||||||||
Veranstaltungsart: | Andere | ||||||||||||||||||||||||||||
Veranstaltungsdatum: | 24 Januar 2023 | ||||||||||||||||||||||||||||
Veranstalter : | Institute for Advanced Simulation, Jülich Supercomputing Centre | ||||||||||||||||||||||||||||
HGF - Forschungsbereich: | Luftfahrt, Raumfahrt und Verkehr | ||||||||||||||||||||||||||||
HGF - Programm: | Luftfahrt | ||||||||||||||||||||||||||||
HGF - Programmthema: | Effizientes Luftfahrzeug | ||||||||||||||||||||||||||||
DLR - Schwerpunkt: | Luftfahrt | ||||||||||||||||||||||||||||
DLR - Forschungsgebiet: | L EV - Effizientes Luftfahrzeug | ||||||||||||||||||||||||||||
DLR - Teilgebiet (Projekt, Vorhaben): | L - Virtuelles Flugzeug und Validierung | ||||||||||||||||||||||||||||
Standort: | Dresden | ||||||||||||||||||||||||||||
Institute & Einrichtungen: | Institut für Softwaremethoden zur Produkt-Virtualisierung > Hochleistungsrechnen | ||||||||||||||||||||||||||||
Hinterlegt von: | Rempke, Arne | ||||||||||||||||||||||||||||
Hinterlegt am: | 07 Feb 2023 13:27 | ||||||||||||||||||||||||||||
Letzte Änderung: | 28 Mai 2024 09:06 |
Nur für Mitarbeiter des Archivs: Kontrollseite des Eintrags