The turbulent mixing modeling for unsteady heat transfer problems in thermal-hydraulics has long been a focus in nuclear engineering community. One of the promising approaches that takes a full advantage of recent advances in HPC is the use of hierarchy of fidelity simulations that are cross-verified and validated in the regimes of interest. In particular, it is beneficial to have a higher fidelity reference simulation that could be used for benchmarking of the faster-turn-around lower-fidelity tools. This paper focuses on Nek5000 LES reference simulations for the OECD/NEA T-junction and Matis benchmarks and for the "SIBERIA" and MAX con-current validation experiments. The CFD capabilities to predict the unsteady heat transfer problems that are associated with non- isothermal flow mixing in pipe flows and related to thermal fatigue of nuclear power plant pipe system have been tested in the 2010 Vattenfall T-junction benchmark. The 2012 KAERI Matis experiment has tested the CFD predictions in a complex geometry of subchannel mixing flow behind a spacer grid. The Nek5000 LES submissions scored highly in the both blind benchmarks. The Novosibirsk's "SIBERIA" experiment employs a novel electrochemical technique for accurate measurement of the wall mean shear and its fluctuations in an annular channel flow with a blockage. The ANL's MAX experiment is designed to study thermal mixing in a wide mixing chamber where two out of a maximum of four adjacent differentially heated jets are injected to impinge on the top wall surface, leading to a thermal striping phenomenon. The resulting complex flow configuration is tracked by a high-resolution particle velocimetry method and thermal imaging. The high-quality benchmark data profiles and high-resolution MAX data provided are used for validating numerical simulation capabilities of Nek5000 for these types of flows and for benchmarking and improvement of lower-fidelity uRANS modeling as a part of hierarchical multi-fidelity approach.