We propose to use Dardel’s high-performance computing resources to develop and assess reduced-order models (ROMs) for large-scale turbulent CFD simulations using Proper Orthogonal Decomposition (POD), Dynamic Mode Decomposition (DMD) and Spectral POD (SPOD). High-fidelity simulations of complex, separated and wall-bounded flows at moderate to high Reynolds numbers generate snapshot datasets with billions of degrees of freedom and thousands of time instances, far beyond the capacity of local workstations for storage, I/O and linear-algebra workloads. On Dardel we will (i) run or post-process large CFD datasets, (ii) construct snapshot matrices distributed across nodes, and (iii) compute POD, DMD and SPOD modes using scalable parallel linear algebra. The goal is to extract energy-optimal and dynamically relevant coherent structures, quantify dominant frequencies and growth/decay rates, and build low-dimensional ROMs suitable for fast prediction and flow-control design. We will benchmark the different modal decompositions in terms of reconstruction error, spectral content, robustness to noise and truncation, and assess their scalability with respect to grid size, number of snapshots and MPI task count. The requested compute will primarily be used for large matrix factorizations (SVD/eigendecompositions), FFT-based SPOD, and high-throughput I/O from CFD outputs. The expected outcome is a validated, scalable workflow for POD/DMD/SPOD on Dardel and a set of ROMs that significantly reduce computational cost while preserving the key physics of the underlying turbulent flows.