The growing availability of GPU-accelerated open-source solvers has boosted the capability of tackling complex single-phase and multiphase turbulent flows by means of direct and large-eddy simulations. GPU-accelerated solvers can leverage the heterogeneous computing architectures that are available in leading high-performance computing centers worldwide, taking advantage of the higher throughput and greater energy efficiency offered by GPUs as compared to CPUs. However, porting CPU-based numerical solvers to GPUs entails many outstanding challenges, such as parallelism exposure, inter-GPU communication, memory allocation constraints, and shared memory limitations. To overcome these challenges, GPU-friendly algorithms, performance portability strategies, and careful selection of computational paradigms and programming languages must be developed. Besides, adaptive mesh refinement and data compression may be integrated to mitigate I-O bottlenecks and enable simulations of more complex geometries on top of the existing requirements imposed by incompressible flows. When compressibility effects become significant, further considerations related to the adoption of high-performance preconditioners and multigrid solvers become crucial for tackling large, sparse linear systems and extending simulations to high-Mach flows. Finally, reduced-precision arithmetic can further enhance performance, energy efficiency, and scalability. In this work, we survey current applications of GPU-accelerated solvers in the broad area of fluid mechanics and turbulence simulations and discuss the main challenges and bottlenecks associated with code porting and optimization. We then conclude our analysis with an outlook on future perspectives for enabling efficient GPU-based exascale computing of turbulence.

As heterogeneous supercomputing architectures leveraging GPUs become increasingly central to high-performance computing (HPC), it is crucial for computational fluid dynamics (CFD) simulations, a de-facto HPC workload, to efficiently utilize such hardware. One of the key challenges of HPC codes is performance portability, i.e. the ability to maintain near-optimal performance across different accelerators. In the context of the REFMAP project, which targets scalable, GPU-enabled multi-fidelity CFD for urban airflow prediction, this paper analyzes the performance portability of SOD2D, a state-of-the-art Spectral Elements simulation framework across AMD and NVIDIA GPU architectures. We first discuss the physical and numerical models underlying SOD2D, highlighting its computational hotspots. Then, we examine its performance and scalability in a multi-level manner, i.e. defining and characterizing an extensive full-stack design space spanning across application, software and hardware infrastructure related parameters. Single-GPU performance characterization across server-grade NVIDIA and AMD GPU architectures and vendor-specific compiler stacks, show the potential as well as the diverse effect of memory access optimizations, i.e. 0.69× - 3.91× deviations in acceleration speedup. Performance variability of SOD2D at scale is further examined on the LUMI multi-GPU cluster, where profiling reveals similar throughput variations, highlighting the limits of performance projections and the need for multi-level, informed tuning.

As heterogeneous supercomputing architectures leveraging GPUs become increasingly central to high-performance computing (HPC), it is crucial for computational fluid dynamics (CFD) simulations to maintain performance portability. In this paper, we examine the performance and scalability of CFD framework SOD2D in a crosslayer manner, i.e. across application, software and hardware infrastructure related parameters. Single-GPU performance characterization across server-grade NVIDIA and AMD GPU architectures and vendor-specific compiler stacks, show the potential as well as the diverse effect of memory access optimizations, i.e. 0.69x - 3.96x deviations in acceleration speedup. Performance variability of SOD2D at scale is then further examined on the LUMI multi-GPU cluster, showcasing analogous diverse effects on throughput, demonstrating the ineffectiveness of adopting performance projections, thus underscoring the importance and necessity of cross-layer informed performance analysis and tuning for multi-GPU configurations.