Hybrid Computing Platform to Enable Complex Multi-Physics DNS/LES from Desktop to Exascale

Abstract

We propose to develop the hybrid CPU-GPU paradigm for unstructured overset grid large-eddy simulation, that promises transformational speedups in computing time. Our goal is to enable exascale simulations on the largest platforms as well as highly affordable simulations on desktop machines. We will develop the capability for our in-house solver MPCUGLES and demonstrate its impact on our current ONR projects. The speedup enabled by the proposed research will significantly impact the utilization of LES for practical problems. Developing a GPU extension for a code as complex as MPCUGLES is not straightforward and involves significant algorithm and data-handling development, and hardware-aware programming. The proposed machine is essential to (i) develop the hybrid accelerated methodology, (ii) evaluate it for relevant problems like crashback, cavitation and hull maneuvering, (iii) explore the potential in detail in terms of what is feasible on machines ranging from desktop machines to leadership class machines. If the promise of the capability being developed is realized, it will enable LES of practical Navy relevant problems to be performed on desktops. Approved for public release.

Document Details

Document Type
DoD Grant Award
Publication Date
Oct 13, 2023
Source ID
N000142312894

Entities

People

  • Krishnan Mahesh

Organizations

  • Board of Regents of the University of Michigan
  • Office of Naval Research
  • United States Navy

Tags

Readers

  • Computational Fluid Dynamics (CFD)
  • Distributed Systems and Data Platform Development
  • Parallel and Distributed Computing.