Multi-GPU Performance Portability Analysis for Scalable CFD Simulations

arXiv.org · January 21, 2026 · ✓ verified

Panagiotis-Eleftherakis and co-authors (REFMAP project) have published an arXiv paper analyzing the performance portability of the SOD2D Spectral Elements CFD framework across AMD and NVIDIA GPUs (arXiv:2601.14159, submitted 20 Jan 2026).

  • Main action: The paper reports a multi-level performance and scalability analysis of SOD2D, including single-GPU characterization and scale tests on the LUMI multi-GPU cluster, documenting 0.69× - 3.91× deviations in acceleration speedup and highlighting variability across vendor-specific compiler stacks and memory-access optimizations.
  • Background and details: The work is produced by a multi-institution team (National Technical University of Athens, KTH, HPE France, TU Delft, University of Michigan) under the REFMAP context; it provides a full-stack design-space characterization (application, software, hardware) and cites profiling results from LUMI; submission date 20 Jan 2026 and DOI https://doi.org/10.48550/arXiv.2601.14159.
Keep reading
Webinar: Canada's Cloud Sovereignty—Where Should the Lines Fall? Information Technology and Innovation Foundation · Jun 09 ANGOTIC 2026 spotlights digital transformation, AI, startups and infrastructure APO Group - Africa · Jun 05 Bank of Italy publishes five new QEF research papers Banca d'Italia · Jun 05 Global data center expansion and implications for electricity consumption Banca d'Italia · Jun 05
Telborg · US Data Centers
Track the US data-center buildout — every day.

Real-time verified news and daily AI-written briefings, built from primary sources — power, grid, permits, land, financing. Start free.

Get Telborg Pro · $189/mo Get the daily briefing — free →