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.

The wake topology behind a wall-mounted square cylinder immersed in a turbulent boundary layer is investigated using high-resolution large-eddy simulations (LES). The boundary-layer thickness at the obstacle location is fixed, with a Reynolds number based on cylinder height ℎ and free-stream velocity 𝑢∞ of 10,000 while the aspect ratio (AR), defined as obstacle height divided by its width, ranges from 1 to 4. The mesh resolution is comparable to DNS standards used for similar wall-mounted obstacles, though with relatively lower Reynolds numbers. The effects of AR on wake structures, turbulence production, and transport are analyzed via Reynolds stresses, anisotropy-invariant maps (AIM), and the turbulent kinetic energy (TKE)budget. In particular, the transition from ‘‘dipole’’ to a ‘‘quadrupole’’ wake is extensively examined as AR increases. With increasing AR, the wake shrinks in both the streamwise and spanwise directions, attributed to the occurrence of the base vortices (AR = 3 and 4). This change in the flow structure also affects the size of the positive-production region that extends from the roof and the flank of the obstacle to the wake core. The AIMs confirm three-dimensional wake features, showing TKE redistribution in all directions (Simonsen and Krogstad, 2005). Stronger turbulence production in AR = 3 and 4 cases highlights the role of tip and base vortices behind the cylinder. The overall aim is to refine the dipole-to-quadrupole transition as a function of AR and accounting for the incoming TBL properties. The novelty relies on proposing the momentum-thickness-based Reynolds number Re𝜃 as a discriminant for assessing TBL effects on turbulent wake structures.

High-resolution simulations have been extensively utilized to analyze the turbulent structures developing around wall-mounted square cylinders immersed in a turbulent boundary layer. While previous studies have demonstrated that parameters, such as the turbulence intensity of the incoming flow and the cylinder aspect ratio, significantly influence flow structures around isolated obstacles, the interaction between multiple obstacles introduces additional complexity. To systematically investigate the physics of this interaction, high-resolution Large-eddy simulations are carried out for two wall-mounted, square cylinders with different heights h1 and h2, and the same width d, in a tandem configuration. The inflow in all cases is a canonical zero-pressure-gradient turbulent boundary layer at a friction Reynolds number ≈180 upstream the leading obstacle. Three configurations are distinguished by increasing obstacle separation, namely, “skimming flow,” “wake interfence,” and “isolated roughness” regimes, in analogy to the flow classification of a building array. While previous studies suggest that these regimes may also qualitatively describe the flow around two identical cylinders, the present paper shows that for cylinders with different heights, the combined effect of the obstacle separation G, and the aspect ratio of the rear cylinder is also critically important. In addition to the mean velocity fields, we examined the vortical motions and the turbulent kinetic energy budget to further reveal how the changes in the vortex dynamics induced by both h2 and the obstacle separation affect the energy exchange from the fluctuation field to the mean flow.

The aviation field is rapidly evolving towards an era where both typical aviation and Unmanned Aicraft Systems are essential and co-exist in the same airspace. This new territory raises important concerns regarding environmental impact, safety and societal acceptance. The RefMap European Project is an initiative that addresses these issues and aims at optimizing air traffic in terms of the environmental footprint in aviation and drone flights. One of RefMap's objectives is the development of powerful deep-learning models that predict urban flow based on extensive CFD simulations. The excessive time requirements of CFD simulations require the computational power of exascale heterogeneous supercomputer clusters. This work presents RefMap's strategy to mitigate simulation to GPU-enabled high-class solvers and further leverage sophisticated autotuning HPC techniques for creating portable high-performance simulations that can efficiently run on any GPU architecture and parallel system.

Prandtl's secondary flows of the second kind generated by laterally varying roughness are studied using the linearised Reynolds-averaged Navier-Stokes approach proposed by Zampino et al. (J. Fluid Mech., vol. 944, 2022, p. A4). The momentum equations are coupled to the Spalart-Allmaras model while the roughness is captured by adapting established strategies for homogeneous roughness to heterogeneous surfaces. Linearisation of the governing equations yields a framework that enables a rapid exploration of the parameter space associated with heterogeneous surfaces, in the limiting case of small spanwise variations of the roughness properties. Channel flow is considered, with longitudinal high- and low-roughness strips arranged symmetrically. By varying the strip width, it is found that linear mechanisms play a dominant role in determining the size and intensity of secondary flows. In this setting, secondary flows may be interpreted as the time-averaged output response of the turbulent mean flow subjected to a steady forcing produced by the wall heterogeneity. In fact, the linear model predicts that secondary flows are most intense when the strip width is about 0.7 times the half-channel height, in excellent agreement with available data. Furthermore, a unified framework to analyse combinations of heterogeneous roughness properties and laterally varying topographies, common in applications, is discussed. Noting that the framework assumes small spanwise variations of the surface properties, two separate secondary-flow-inducing source mechanisms are identified, i.e. the lateral variation of the virtual origin from which the turbulent structure develops and the lateral variation of the streamwise velocity slip, capturing the acceleration/deceleration perceived by the bulk flow over troughs and crests of non-planar topographies.