Results for “mixed precision solvers” — Page 2

DOE OSTI2023

Large-Scale Materials Modeling at Quantum Accuracy: Ab Initio Simulations of Quasicrystals and Interacting Extended Defects in Metallic Alloys

Ab initio electronic-structure has remained dichotomous between achievable accuracy and length-scale. Quantum many-body (QMB) methods realize quantum accuracy but fail to scale. Density functional theory (DFT) scales favorably but remains far from quantum accuracy. We present a framework that breaks this dichotomy by use of three interconnected modules: (i) invDFT: a methodological advance in inverse DFT linking QMB methods to DFT; (ii) MLXC: a machine-learned density functional trained with invDFT data, commensurate with quantum accuracy; (iii) DFT-FE-MLXC: an adaptive higher-order spectral finite-element (FE) based DFT implementation that integrates MLXC with efficient solver strategies and HPC innovations in FE-specific dense linear algebra, mixed-precision algorithms, and asynchronous compute-communication. Furthermore, we demonstrate a paradigm shift in DFT that not only provides an accuracy commensurate with QMB methods in ground-state energies, but also attains an unprecedented performance of 659.7 PFLOPS (43.1% peak FP64 performance) on 619,124 electrons using 8,000 GPU nodes of Frontier supercomputer.

density functional theory↗

DOE OSTI2023

Combining Sparse Approximate Factorizations with Mixed-precision Iterative Refinement

The standard LU factorization-based solution process for linear systems can be enhanced in speed or accuracy by employing mixed-precision iterative refinement. Most recent work has focused on dense systems. We investigate the potential of mixed-precision iterative refinement to enhance methods for sparse systems based on approximate sparse factorizations. In doing so, we first develop a new error analysis for LU- and GMRES-based iterative refinement under a general model of LU factorization that accounts for the approximation methods typically used by modern sparse solvers, such as low-rank approximations or relaxed pivoting strategies. We then provide a detailed performance analysis of both the execution time and memory consumption of different algorithms, based on a selected set of iterative refinement variants and approximate sparse factorizations. Our performance study uses the multifrontal solver MUMPS, which can exploit block low-rank factorization and static pivoting. We evaluate the performance of the algorithms on large, sparse problems coming from a variety of real-life and industrial applications showing that mixed-precision iterative refinement combined with approximate sparse factorization can lead to considerable reductions of both the time and memory consumption.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2022

Compressed basis GMRES on high-performance graphics processing units

Krylov methods provide a fast and highly parallel numerical tool for the iterative solution of many large-scale sparse linear systems. To a large extent, the performance of practical realizations of these methods is constrained by the communication bandwidth in current computer architectures, motivating the investigation of sophisticated techniques to avoid, reduce, and/or hide the message-passing costs (in distributed platforms) and the memory accesses (in all architectures). This article leverages Ginkgo’s memory accessor in order to integrate a communication-reduction strategy into the (Krylov) GMRES solver that decouples the storage format (i.e., the data representation in memory) of the orthogonal basis from the arithmetic precision that is employed during the operations with that basis. Given that the execution time of the GMRES solver is largely determined by the memory accesses, the cost of the datatype transforms can be mostly hidden, resulting in the acceleration of the iterative step via a decrease in the volume of bits being retrieved from memory. Together with the special properties of the orthonormal basis (whose elements are all bounded by 1), this paves the road toward the aggressive customization of the storage format, which includes some floating-point as well as fixed-point formats with mild impact on the convergence of the iterative process. We develop a high-performance implementation of the “compressed basis GMRES” solver in the Ginkgo sparse linear algebra library using a large set of test problems from the SuiteSparse Matrix Collection. We demonstrate robustness and performance advantages on a modern NVIDIA V100 graphics processing unit (GPU) of up to 50% over the standard GMRES solver that stores all data in IEEE double-precision.

97 MATHEMATICS AND COMPUTING↗

NASA NTRS2024

Global Optimization via Quadratic Disjunctive Programming for Water Networks Design with Energy Recovery

Generalized disjunctive programming (GDP) models with bilinear and concave constraints, often seen in water network design, are challenging optimization problems. This work proposes quadratic and piecewise linear approximations for nonlinear terms to reformulate GDP models into quadratic GDP (QGDP) models that suitable solvers may solve more efficiently. We illustrate the benefits of the quadratic reformulation with a water treatment network design problem in which nonconvexities arise from bilinear terms in the mixers’ mass balances and concave investment cost functions of treatment units. Given the similarities with water network design problems, we suggest quadratic approximation for the GDP model for the optimal design of a large-scale reverse electrodialysis (RED) process. This power technology can recover energy from salinity differences between by-product streams of the water sector, such as desalination brine mixed with regenerated wastewater effluents. The solver Gurobi excels in handling QGDP problems, but weighing the problem’s precision and tractability balance is crucial. The piecewise linear approximation yields more accurate, yet larger QGDP models that may require longer optimization times in large-scale process synthesis problems.

Water Networks↗

DOE OSTI2025

Characterizing GPU Energy Usage in Exascale-Ready Portable Science Applications

We characterize the GPU energy usage of two widely adopted exascale-ready applications representing two classes of particle and mesh solvers: (i) QMCPACK, a quantum Monte Carlo package, and (ii) AMReX-Castro, an adaptive mesh astrophysical code. We analyze power, temperature, utilization, and energy traces from double-/single (mixed)-precision benchmarks on NVIDIA’s A100 and H100 and AMD’s MI250X GPUs using queries in NVML and rocm_smi_lib, respectively. We explore application-specific metrics to provide insights on energy vs. performance trade-offs. Our results suggest that mixed-precision energy savings range between 6–25% on QMCPACK and 45% on AMReX-Castro. Also, we found gaps in the AMD tooling used on Frontier GPUs that need to be understood, while query resolutions on NVML have little variability between 1 ms-1 s. Overall, application level knowledge is crucial to define energy-cost/science-benefit opportunities for the codesign of future supercomputer architectures in the post-Moore era.

Godoy, William [ORNL] (ORCID:0000000225905178)↗

NASA NTRSDate not supplied

LES-Based Analysis of Film Cooling in a High-Pressure Turbine Rotor Including Effects of Purge and Tip Flow

Accurate modeling of coolant airflows, which form protective films over turbine blades, is essential for designing fuel-efficient and environmentally sustainable gas turbine engines. Excessive coolant reduces thermal efficiency, while insufficient coolant leads to blade overheating and causes damage. Therefore, precise prediction of flow field interactions with cooling air is critical for optimizing turbine performance. This study numerically investigates the cooling effectiveness of purge and film cooling flows within a high-pressure turbine (HPT) rotor using Large Eddy Simulation (LES). The study utilizes NASA Glenn Research Center’s Glenn-HT solver. The simulation models ethe conditions of the Penn State University START rotating rig. A high-fidelity structured mesh comprising up to 800 million cells is employed to resolve high-Reynolds number flow (Re ≈ 350,000) and to capture intricate secondary flow structures, including tip leakage and purge-induced vortices. Film cooling effectiveness computations are highly sensitive to boundary conditions at the cooling holes and to grid resolution. Even with well-resolved grids and included plena, strong mixing challenges traditional eddy viscosity models. To address this, a simplified configuration is simulated: a truncated row of shaped holes on the suction side near the leading edge and a row on the pressure side, both fed from internal plena while the purge slot and tip clearance are also modeled. Two isothermal LES cases are conducted at two distinct wall temperatures, which yield the adiabatic wall temperature and the heat transfer coefficient. The definition and means of computation of the effectiveness is discussed in this paper. The simulations reveal detailed three-dimensional unsteady flow features, including coherent vortical structures and secondary flows originating from the purge cavity. Film cooling effectiveness and Nusselt number distributions are presented for both the blade surface and tip, highlighting regions of elevated heat transfer and complex thermal behavior. These findings underscore the importance of high-resolution LES and realistic boundary conditions in capturing the dynamics of purge and film cooling, offering valuable insights for improving turbine blade design and thermal management strategies.

Secondary Flows↗

NASA NTRSDate not supplied

A GPU-based Approach for Turbomachinery Application

Accurate modeling of coolant airflows, which form protective films over turbine blades, is essential for designing fuel-efficient and environmentally sustainable gas turbine engines. Excessive coolant reduces thermal efficiency, while insufficient coolant leads to blade overheating and causes damage. Therefore, precise prediction of flow field interactions with cooling air is critical for optimizing turbine performance. This study numerically investigates the cooling effectiveness of purge and film cooling flows within a high-pressure turbine (HPT) rotor using Large Eddy Simulation (LES). The study utilizes NASA Glenn Research Center’s Glenn-HT solver. The simulation models ethe conditions of the Penn State University START rotating rig. A high-fidelity structured mesh comprising up to 800 million cells is employed to resolve high-Reynolds number flow (Re ≈ 350,000) and to capture intricate secondary flow structures, including tip leakage and purge-induced vortices. Film cooling effectiveness computations are highly sensitive to boundary conditions at the cooling holes and to grid resolution. Even with well-resolved grids and included plena, strong mixing challenges traditional eddy viscosity models. To address this, a simplified configuration is simulated: a truncated row of shaped holes on the suction side near the leading edge and a row on the pressure side, both fed from internal plena while the purge slot and tip clearance are also modeled. Two isothermal LES cases are conducted at two distinct wall temperatures, which yield the adiabatic wall temperature and the heat transfer coefficient. The definition and means of computation of the effectiveness is discussed in this paper. The simulations reveal detailed three-dimensional unsteady flow features, including coherent vortical structures and secondary flows originating from the purge cavity. Film cooling effectiveness and Nusselt number distributions are presented for both the blade surface and tip, highlighting regions of elevated heat transfer and complex thermal behavior. These findings underscore the importance of high-resolution LES and realistic boundary conditions in capturing the dynamics of purge and film cooling, offering valuable insights for improving turbine blade design and thermal management strategies.

Gas Turbine↗

NASA NTRS2006

Unstructured CFD and Noise Prediction Methods for Propulsion Airframe Aeroacoustics

Using unstructured mesh CFD methods for Propulsion Airframe Aeroacoustics (PAA) analysis has the distinct advantage of precise and fast computational mesh generation for complex propulsion and airframe integration arrangements that include engine inlet, exhaust nozzles, pylon, wing, flaps, and flap deployment mechanical parts. However, accurate solution values of shear layer velocity, temperature and turbulence are extremely important for evaluating the usually small noise differentials of potential applications to commercial transport aircraft propulsion integration. This paper describes a set of calibration computations for an isolated separate flow bypass ratio five engine nozzle model and the same nozzle system with a pylon. These configurations have measured data along with prior CFD solutions and noise predictions using a proven structured mesh method, which can be used for comparison to the unstructured mesh solutions obtained in this investigation. This numerical investigation utilized the TetrUSS system that includes a Navier-Stokes solver, the associated unstructured mesh generation tools, post-processing utilities, plus some recently added enhancements to the system. New features necessary for this study include the addition of two equation turbulence models to the USM3D code, an h-refinement utility to enhance mesh density in the shear mixing region, and a flow adaptive mesh redistribution method. In addition, a computational procedure was developed to optimize both solution accuracy and mesh economy. Noise predictions were completed using an unstructured mesh version of the JeT3D code.

Pao, S. Paul↗

NASA NTRSDate not supplied

Scheduling Position, Navigation and Time Service Requests from Non-dedicated Lunar Constellations

This paper presents a centralized scheduler that satisfies user requests for Position, Navigation, and Time (PNT) services from an ad-hoc, non-dedicated orbital constellation around the Moon. Traditional, dedicated GNSS networks provide service 24/7, which allows users to acquire localization services at-will. For ad-hoc networks, a coordinated schedule is needed to ensure Quality of Service (QoS) guarantees for user localization, while satisfying non-dedicated assets’ usage constraints. This scheduler bridges this coordination gap by leveraging Mixed Integer-Linear Programming (MILP) to schedule this “as-needed” localization service while respecting the constraints on each asset. In upcoming decades there is expected to be a substantial increase in Lunar missions. Many of these missions will feature low-cost surface assets near the moon’s polar regions and small-sat science missions in orbit. Most missions need PNT capabilities to ensure safe operations and meet their science objectives, but low-cost missions may not be able to support the large power, mass, and weight that a weak GNSS or DSN based navigation solution would entail. Asset localization has been demonstrated using a decentralized extended Kalman Filter (DEKF) in the previously presented Lunar Autonomous PNT System (LAPS). Within the LAPS simulation environment, a module has been developed to generate the coordinated user-asset schedules described above; this Service Scheduler Module (SSM) allows for complete end-to-end testing of the entire system. Within SSM, a user service request consists of a location on the Lunar surface, a cumulative service duration, and a window in which service must occur. SSM takes as input these requests and the LAPS-predicted positional degree of precision as the QoS for each available set of orbital assets. A simple, baseline MILP model is formulated to provide the highest-precision service balanced across all requests. To reflect the non-dedicated nature of the constellation, this baseline model is augmented with additional asset-specific load capacity constraints or availability constraints. The load capacity constraints limit total time spent providing service, and the availability constraints reflect blockout times or availability windows when the assets are not otherwise occupied. SSM outputs two schedules: the user schedule to indicate their service times and expected QoS, and a satellite schedule to be transmitted to the orbiting constellation, describing when each non-dedicated asset provides PNT service. SSM is predominantly implemented in MATLAB and allows the use of any MILP solver to generate the resulting schedules. This paper describes the SSM - LAPS interface, how the output of LAPS is used to construct the MILP, and how SSM provides user localization service while satisfying constraints. It will also demonstrate the tool’s flexibility for formulating schedules for the end user and the constellation, focusing on scenarios that match real-world proposed missions. It will detail how SSM can be used to compare the addition of load capacity constraints, satellite availability constraints, and QoS guarantees for the users. Finally, we describe how SSM can be used to support the design of the ad-hoc constellation itself. The resulting integrated capability will support the design of future ad-hoc Lunar PNT networks, enabling high-quality, low-cost Lunar exploration

Swarm↗