Towards GPU-Accelerated Flow Solvers for Planetary Entry

The DPLR and LAURA computational fluid dynamic (CFD) solvers are the foundation of NASA’s capability to estimate aerodynamic heating for manned and un-manned entry, descent and landing (EDL) systems. However, NASA’s technology roadmap calls for planetary entry concepts, most notably Supersonic Retro Propulsion (SRP), which require extremely complex aeroheating simulations that cannot be completed in design-relevant timespans using current supercomputers. While deployment of exascale-class supercomputers can alleviate this bottleneck, leadership-class machines increasingly rely on many-core accelerators, e.g. Graphics Processing Units (GPUs), to provide compute capacity. Recent work by NASA’s FUN3D team on the DoE machine Summit (the world’s fastest computer) demonstrated that a single nVidia Volta GPU matches the throughput of ~11 modern Intel Skylake CPUs. These GPUs are now being deployed to NASA’s Pleiades supercomputer, but DPLR and LAURA cannot exploit these devices because they fundamentally unable to run on many-core architectures. In collaboration with the Langley Aerothermodynamics branch, this effort proposes to migrate and demonstrate key EDL-relevant compute kernels on Volta GPUs using Sandia’s Kokkos Performance Portability Library. Academia, industry and government have all invested heavily in many-core accelerated CFD solvers. Notable examples are PyFR from Imperial College London, the Ansys Fluent commercial solver, and NASA’s FUN3D solver. However, nearly all previous work has focused on perfect gas flow models, which have different algorithmic and resource requirements compared to real gas models. Sandia’s SPARC solver is the only notable GPU-ready real gas solver, but published work only considers relatively simple chemical models (5 species air in thermal equilibrium) and did not assess scaling of this model on the Volta GPU. This proposal expands the state of the art by investigating models and algorithms needed for orbital- and super-orbital planetary entry on nVidia’s Volta GPUs, which are now available on NASA’s Pleiades supercomputer. Furthermore, in contrast to FUN3D, which uses the nVidiaspecific CUDA programming language, this work is based on Sandia’s Kokkos library, which allows running standard C++ code on multiple compute architectures with minimal modification. As such, this effort leverages DoE investment in many-core compute libraries to reduce development time, avoid vendor/platform lock-in, and improve maintainability of our code.

FUN3D has shown ~30x reduction in time-to-solution using GPUs. However, this result did not model the gas in thermochemical nonequilibrium as required for planetary entry. Sandia’s SPARC code has demonstrated a performant 5-species reacting air model, but also neglected nonequilibrium thermodynamic effects.

This is a mature technology (TRL 7+) — the realistic path in is usually NASA's Technology Transfer Program: licensing an existing NASA patent, or a Space Act Agreement to use NASA facilities/expertise directly. NASA also runs a startup licensing program with no upfront fee for companies formed to commercialize a specific NASA technology.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.