Search NASA⌕ Search

SEARCH · Search NASA

Results for “Parallelization efficiency”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,333 records · Page 74

MoE-Inference-Bench: Performance Evaluation of Mixture of Expert Large Language and Vision Models

Mixture of Experts (MoE) models have enabled the scaling of Large Language Models (LLMs) and Vision Language Models (VLMs) by achieving massive parameter counts while maintaining computational efficiency. However, MoEs introduce several inference-time challenges, including load imbalance across experts and the additional routing computational overhead. To address these challenges and fully harness the benefits of MoE, a systematic evaluation of hardware acceleration techniques is essential. We present MoE-Inference-Bench, a comprehensive study to evaluate MoE performance across diverse scenarios. We analyze the impact of batch size, sequence length, and critical MoE hyperparameters such as FFN dimensions and number of experts on throughput. We evaluate several optimization techniques on Nvidia H100 GPUs, including pruning, Fused MoE operations, speculative decoding, quantization, and various parallelization strategies. Our evaluation includes MoEs from the Mixtral, DeepSeek, OLMoE and Qwen families. The results reveal performance differences across configurations and provide insights for the efficient deployment of MoEs.

Chitty-Venkata, Krishna Teja↗

Cross-scale efficient tensor contractions for coupled cluster computations through multiple programming model backends

Coupled-cluster methods provide highly accurate models of molecular structure through explicit numerical calculation of tensors representing the correlation between electrons. These calculations are dominated by a sequence of tensor contractions, motivating the development of numerical libraries for such operations. While based on matrix–matrix multiplication, these libraries are specialized to exploit symmetries in the molecular structure and in electronic interactions, and thus reduce the size of the tensor representation and the complexity of contractions. The resulting algorithms are irregular and their parallelization has been previously achieved via the use of dynamic scheduling or specialized data decompositions. We introduce our efforts to extend the Libtensor framework to work in the distributed memory environment in a scalable and energy-efficient manner. We achieve up to 240× speedup compared with the optimized shared memory implementation of Libtensor. We attain scalability to hundreds of thousands of compute cores on three distributed-memory architectures (Cray XC30 and XC40, and IBM Blue Gene/Q), and on a heterogeneous GPU-CPU system (Cray XK7). As the bottlenecks shift from being compute-bound DGEMM's to communication-bound collectives as the size of the molecular system scales, we adopt two radically different parallelization approaches for handling load-imbalance, tasking and bulk synchronous models. Nevertheless, we preserve a unified interface to both programming models to maintain the productivity of computational quantum chemists.

97 MATHEMATICS AND COMPUTING↗

Toucan: A performance portable, scalable implementation of the DECA algorithm

In the field of additive manufacturing (AM), cellular automata (CA) is extensively used to simulate microstructural evolution during solidification. However, while traditional CA approaches are relatively fast, they still require a substantial number of time steps, are limited to moderate volumes, and are relatively difficult to improve through parallelism due to the highly localized nature of the solidification front. Here, to address these issues of time to solution and load balancing, we introduce Toucan, a parallel, performance-portable, and scalable code written in C++ with the Kokkos library that leverages the discrete event inspired cellular automata (DECA) algorithm to perform parallel-in-time (PinT) grain growth simulations. Toucan effectively mitigates load balancing issues by distributing the computational workload more evenly across processors, enhancing scalability and efficiency. We conduct both strong and weak scaling studies on up to 64 GPUs on the Frontier supercomputer, demonstrating that Toucan significantly outperforms the current state-of-the-art, time-stepped CA code, ExaCA, on both single and multi-GPU simulations. Even in AM-specific weak scaling scenarios, Toucan maintains near-ideal scaling, in contrast to the linear increase observed with ExaCA due to the moving laser raster pattern. This study highlights Toucan’s potential to transform microstructural simulations in AM by radically improving both efficiency and scalability over existing methods.

36 MATERIALS SCIENCE↗

SNAPPy: A snakemake pipeline for scalable HIV-1 subtyping by phylogenetic pairing

Abstract Human immunodeficiency virus 1 (HIV-1) genome sequencing is routinely done for drug resistance monitoring in hospitals worldwide. Subtyping these extensive datasets of HIV-1 sequences is a critical first step in molecular epidemiology and evolution studies. The clinical relevance of HIV-1 subtypes is increasingly recognized. Several studies suggest subtype-related differences in disease progression, transmission route efficiency, immune evasion, and even therapeutic outcomes. HIV-1 subtyping is mainly done using web-servers. These tools have limitations in scalability and potential noncompliance with data protection legislation. Thus, the aim of this work was to develop an efficient method for large-scale local HIV-1 subtyping. We designed SNAPPy: a snakemake pipeline for scalable HIV-1 subtyping by phylogenetic pairing. It contains several tasks of phylogenetic inference and BLAST queries, which can be executed sequentially or in parallel, taking advantage of multiple-core processing units. Although it was built for subtyping, SNAPPy is also useful to perform extensive HIV-1 alignments. This tool facilitates large-scale sequence-based HIV-1 research by providing a local, resource efficient and scalable alternative for HIV-1 subtyping. It is capable of analyzing full-length genomes or partial HIV-1 genomic regions (GAG, POL, and ENV) and recognizes more than ninety circulating recombinant forms. SNAPPy is freely available at: https://github.com/PMMAraujo/snappy/releases.

Araújo, Pedro M. M.↗

Beam loss mechanisms in the PIP-II linac and beam transfer line at Fermilab

Beam loss in high-intensity H- linacs, such as the PIP-II linac at Fermilab, is a critical challenge that requires comprehensive study and understanding to ensure efficient and safe operation. This study explores the various beam loss mechanisms encountered in the PIP-II linac and its beam transfer line, drawing parallels from other high-intensity H- linacs. Key loss mechanisms include residual gas stripping, where H- ions interact with residual gas molecules leading to electron detachment; field stripping, caused by the interaction of H- ions with magnetic fields; and intra-beam stripping, resulting from interactions within the beam itself. Beam halo formation, particularly due to Twiss function mismatch, is another significant source of beam loss, which can be exacerbated by Landau damping mechanisms. Adhering to the 1 W/m loss criterion is essential to maintain hands-on maintenance capability and ensure the longevity of the accelerator components. By understanding these mechanisms and implementing targeted mitigation strategies, the PIP-II linac can achieve its design goals while maintaining safe and efficient operations.

43 PARTICLE ACCELERATORS↗

Transfer Line at Fermilab

Beam loss in high-intensity H- linacs, such as the PIP-II linac at Fermilab, is a critical challenge that requires comprehensive study and understanding to ensure efficient and safe operation. This study explores the various beam loss mechanisms encountered in the PIP-II linac and its beam transfer line, drawing parallels from other high-intensity H- linacs. Key loss mechanisms include residual gas stripping, where H- ions interact with residual gas molecules leading to electron detachment; field stripping, caused by the interaction of H- ions with magnetic fields; and intra-beam stripping, resulting from interactions within the beam itself. Beam halo formation, particularly due to Twiss function mismatch, is another significant source of beam loss, which can be exacerbated by Landau damping mechanisms. Adhering to the 1 W/m loss criterion is essential to maintain hands-on maintenance capability and ensure the longevity of the accelerator components. By understanding these mechanisms and implementing targeted mitigation strategies, the PIP-II linac can achieve its design goals while maintaining safe and efficient operations.

Pathak, Abhishek↗

The role of the electron convection term for the parallel electric field and electron acceleration in MHD simulations

There has been a great concern about the origin of the parallel electric field in the frame of fluid equations in the auroral acceleration region. This paper proposes a new method to simulate magnetohydrodynamic (MHD) equations that include the electron convection term and shows its efficiency with simulation results in one dimension. We apply a third-order semi-discrete central scheme to investigate the characteristics of the electron convection term including its nonlinearity. At a steady state discontinuity, the sum of the ion and electron convection terms balances with the ion pressure gradient. We find that the electron convection term works like the gradient of the negative pressure and reduces the ion sound speed or amplifies the sound mode when parallel current flows. The electron convection term enables us to describe a situation in which a parallel electric field and parallel electron acceleration coexist, which is impossible for ideal or resistive MHD.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Vectorization of Random Number Generation and Reproducibility of Concurrent Particle Transport Simulation

Efficient random number generation with high quality statistical properties and exact reproducibility of Monte Carlo simulations are important requirements in many areas of computational science. VecRNG is a package providing pseudo-random number generation (pRNG) in the context of a new library VecMath. This library bundles up several general-purpose mathematical utilities, data structures, and algorithms having both SIMD and SIMT (GPUs) support based on VecCore.~Several state-of-the-art RNG algorithms are implemented as kernels supporting parallel generation of random numbers in scalar, vector, and Cuda workflows. In this report, we will present design considerations, implementation details, and computing performance of parallel pRNG engines on both CPU and GPU. Reproducibility of propagating multiple particles in parallel for HEP event simulation is demonstrated, using GeantV-based examples, for both sequential and fine-grain track-level concurrent simulation wor kflows. Strategies for efficient uses of vectorized pRNG and non-overlapping streams of random number sequences in concurrent computing environments is discussed as well.

Jun, S. Y.↗

Sludge Settling Rate Observations and Projections at the Savannah River Site - 13238

Since 2004, sludge batches have included a high percentage of stored sludge generated from the H- modified (HM) process. The slow-settling nature of HM sludge means that the settling is often the major part of the washing tank quiescent period between required pump runs to maintain flammability control. Reasonable settling projections are needed to wash soluble salts from sludge in an efficient manner, to determine how much sludge can be washed in a batch within flammability limits, and to provide composition projections for batch qualification work done in parallel with field preparation. Challenges to providing reasonably accurate settling projections include (1) large variations in settling behavior from tank-to-tank, (2) accounting for changing initial concentrations, sludge masses, and combinations of different sludge types, (3) changing the settling behavior upon dissolving some sludge compounds, and (4) sludge preparation schedules that do not allow for much data collection for a particular sludge before washing begins. Scaling from laboratory settling tests has provided inconsistent results. Several techniques have been employed to improve settling projections and therefore the overall batch preparation efficiency. Before any observations can be made on a particular sludge mixture, projections can only be made based on historical experience with similar sludge types. However, scaling techniques can be applied to historical settling models to account for different sludge masses, concentrations, and even combinations of types of sludge. After sludge washing/settling cycles begin, the direct measurement of the sludge height, once generally limited to a single turbidity meter measurement per settle period, is now augmented by examining the temperature profile in the settling tank, to help determine the settled sludge height over time. Recently, a settling model examined at PNNL [1,2,3] has been applied to observed thermocouple and turbidity meter readings to quickly provide settling correlations to project settled heights for other conditions. These tools improve the accuracy and adaptability of short and mid-range planning for sludge batch preparation. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Optimization of multigrid based elliptic solver for large scale simulations in the FLASH code: OPTIMIZATION OF MULTIGRID BASED ELLIPTIC SOLVER IN THE FLASH CODE

FLASH is a multiphysics multiscale adaptive mesh refinement (AMR) code originally designed for simulation of reactive flows often found in Astrophysics. With its wide user base and flexible applications configuration capability, FLASH has a dual task of maintaining scalability and portability in all its solvers. The scalability of fully explicit solvers in the code is tied very closely to that of the underlying mesh. Others such as the Poisson solver based on a multigrid method have more complex scaling behavior. Multigrid methods suffer from processor starvation and dominating communication costs at coarser grids with increase in the number of processors. In this paper, we propose a combination of uniform grid mesh with AMR mesh, and the merger of two different sets of solvers to overcome the scalability limitation of the Poisson solver in FLASH. The principal challenge in the proposed merger is the efficiency of the communication algorithm to map the mesh back and forth between uniform grid and AMR. We present two different parallel mapping algorithms and also discuss results from performance studies of the two implementations. Copyright © 2012 John Wiley & Sons, Ltd.

Computer Science↗

Experiences with string matching on the Fermi Architecture

String matching is at the core of many real-world applications, such as security, bioinformatic, data mining. All these applications requires the ability to match always growing data sets against large dictionaries effectively, fastly and possibly in real time. Unfortunately, string matching is a computationally intensive procedure which poses significant challenges on current software and hardware implementations. Graphic Processing Units (GPU) have become an interesting target for such high-throughput applications, but the algorithms and the data structures need to be redesigned to be parallelized and adapted to the underlining hardware, coping with the limitations imposed by these architectures. In this paper we present an efficient implementation of the Aho-Corasick string matching algorithm on GPU, showing how we progressively redesigned the algorithm and the data structures to fit on the architecture. We then evaluate the implementation on single and multiple Tesla C2050 (T20 ``Fermi'' based) boards, comparing them to the previous Tesla C1060 (T10 based) solutions and equivalent multicore implementations on x86 CPUs. We discuss the various tradeoffs of the different architectures.

99 GENERAL AND MISCELLANEOUS↗

DIRAQ: scalable in situ data- and resource-aware indexing for optimized query performance

Scientific data analytics in high-performance computing environments has been evolving along with the advancement of computing capabilities. With the onset of exascale computing, the increasing gap between compute performance and I/O bandwidth has rendered the traditional post-simulation processing a tedious process. Despite the challenges due to increased data production, there exists an opportunity to benefit from "cheap" computing power to perform query-driven exploration and visualization during simulation time. To accelerate such analyses, applications traditionally augment, post-simulation, raw data with large indexes, which are then repeatedly utilized for data exploration. However, the generation of current state-of-the-art indexes involves a compute- and memory-intensive processing, thus rendering them inapplicable in an in situ context. In this paper we propose DIRAQ, a parallel in situ, in network data encoding and reorganization technique that enables the transformation of simulation output into a query-efficient form, with negligible runtime overhead to the simulation run. DIRAQ's effective core-local, precision-based encoding approach incorporates an embedded compressed index that is 3---6 $$\times $$ ? smaller than current state-of-the-art indexing schemes. Its data-aware index adjustmentation improves performance of group-level index layout creation by up to 35 % and reduces the size of the generated index by up to 27 %. Moreover, DIRAQ's in network index merging strategy enables the creation of aggregated indexes that speed up spatial-context query responses by up to $$10\times $$ 10 ? versus alternative techniques. DIRAQ's topology-, data-, and memory-aware aggregation strategy results in efficient I/O and yields overall end-to-end encoding and I/O time that is less than that required to write the raw data with MPI collective I/O.

Computer Science↗

A fast, efficient direct slicing method for slender member structures

We describe a method for quickly and efficiently slicing structures consisting of a large number of slender members called struts connected at node positions called joints. Previous research on periodic lattice structures shows these structures are highly mechanically efficient with exceptionally high stiffness- and strength-to-weight ratios. Additive manufacturing technologies allow the construction slender member structures with complicated macroscale shapes. These structures could consist of thousands or millions of geometric primitives describing the struts. Structures with large numbers of geometric objects cause the conventional methods for manipulating, storing, and slicing the geometry of these parts via STL files to be highly inefficient. This work describes an alternate design process for slender member structures using efficient methods for manipulating, storing, and slicing the geometry of the part. These new methods, in particular a fast, efficient direct slicing method, enable printing slender member structures with over one hundred thousand struts. The slicing algorithm is nearly perfectly parallel so it could extend to handle structures with over one million struts, helping to facilitate the adoption of slender member structures for engineering-scale applications.

36 MATERIALS SCIENCE↗

Estimating the randomness of quantum circuit ensembles up to 50 qubits

Random quantum circuits have been utilized in the contexts of quantum supremacy demonstrations, variational quantum algorithms for chemistry and machine learning, and blackhole information. The ability of random circuits to approximate any random unitaries has consequences on their complexity, expressibility, and trainability. To study this property of random circuits, we develop numerical protocols for estimating the frame potential, the distance between a given ensemble and the exact randomness. Our tensor-network-based algorithm has polynomial complexity for shallow circuits and is high-performing using CPU and GPU parallelism. We study 1. local and parallel random circuits to verify the linear growth in complexity as stated by the Brown–Susskind conjecture, and; 2. hardware-efficient ansätze to shed light on its expressibility and the barren plateau problem in the context of variational algorithms. Our work shows that large-scale tensor network simulations could provide important hints toward open problems in quantum information science.

97 MATHEMATICS AND COMPUTING↗

Membrane-less microfiltration using inertial microfluidics

Microfiltration is a ubiquitous and often crucial part of many industrial processes, including biopharmaceutical manufacturing. Yet, all existing filtration systems suffer from the issue of membrane clogging, which fundamentally limits the efficiency and reliability of the filtration process. Herein, we report the development of a membrane-less microfiltration system by massively parallelizing inertial microfluidics to achieve a macroscopic volume processing rates (~ 500 mL/min). We demonstrated the systems engineered for CHO (10–20 μm) and yeast (3–5 μm) cells filtration, which are two main cell types used for large-scale bioreactors. Our proposed system can replace existing filtration membrane and provide passive (no external force fields), continuous filtration, thus eliminating the need for membrane replacement. This platform has the desirable combinations of high throughput, low-cost, and scalability, making it compatible for a myriad of microfiltration applications and industrial purposes.

42 ENGINEERING↗

Fabrication and long-wavelength characterization of neat and chemically modified graphene

Graphene, a single- or several layer-thick carbon, attracts significant research activity because of its exceptional material properties. Graphene is a promising material for optoelectronic applications. Neat graphene demonstrates potential as a material for long wavelength photodetectors working at elevated temperatures. Chemical modification of graphene opens up many new applications of this material in electronics, in new composite materials, and in new catalysts for different chemical processes. Chemical vapor deposition-grown large-area graphene can be successfully modified with the creation of benzyne attachments. The investigation of microwave properties is an important part of graphene research. Two variants of near-field long wavelength microscopy were found efficient with graphene. Measurements with a probe formed by an electrically open end of a 4 GHz half-lambda parallel-strip transmission line resonator allow the implementation of an electrodynamic model of graphene microwave impedance. The results of near-field scanning superconducting quantum interference device (SQUID) RF microscopy of graphite and graphene at 200 MHz shed light on mechanisms of AC graphene response: screening currents induced in graphene by an external RF magnetic field tend to localize near structural defects.

36 MATERIALS SCIENCE↗

Exponential time differencing methods with Chebyshev collocation for polymers confined by interacting surfaces

We present a fast and accurate numerical method for the self-consistent field theory calculations of confined polymer systems. It introduces an exponential time differencing method (ETDRK4) based on Chebyshev collocation, which exhibits fourth-order accuracy in temporal domain and spectral accuracy in spatial domain, to solve the modified diffusion equations. Similar to the approach proposed by Hur et al. [Macromolecules 45, 2905 (2012)], non-periodic boundary conditions are adopted to model the confining walls with or without preferential interactions with polymer species, avoiding the use of surface field terms and the mask technique in a conventional approach. The performance of ETDRK4 is examined in comparison with the operator splitting methods with either Fourier collocation or Chebyshev collocation. Numerical experiments show that our exponential time differencing method is more efficient than the operator splitting methods in high accuracy calculations. This method has been applied to diblock copolymers confined by two parallel flat surfaces.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Integrated modeling applications for tokamak experiments with OMFIT

We developed one modeling framework for integrated tasks (OMFIT) is a comprehensive integrated modeling framework which will enable physics codes to interact in complicated workflows, and support scientists at all stages of the modeling cycle. The OMFIT development follows a unique bottom-up approach, where the framework design and capabilities organically evolve to support progressive integration of the components that are required to accomplish physics goals of increasing complexity. OMFIT provides a workflow for easily generating full kinetic equilibrium reconstructions that are constrained by magnetic and motional Stark effect measurements, and kinetic profile information that includes fast-ion pressure modeled by a transport code. It was found that magnetic measurements can be used to quantify the amount of anomalous fast-ion diffusion that is present in DIII-D discharges, and provide an estimate that is consistent with what would be needed for transport simulations to match the measured neutron rates. OMFIT was used to streamline edge-stability analyses, and evaluate the effect of resonant magnetic perturbation (RMP) on the pedestal stability, which have been found to be consistent with the experimental observations. Moreover, the development of a five-dimensional numerical fluid model for estimating the effects of the interaction between magnetohydrodynamic (MHD) and microturbulence, and its systematic verification against analytic models was also supported by the framework. OMFIT was used for optimizing an innovative high-harmonic fast wave system proposed for DIII-D. For a parallel refractive index n(parallel to) > 3, the conditions for strong electron-Landau damping were found to be independent of launched n(parallel to) and poloidal angle. OMFIT has been the platform of choice for developing a neural-network based approach to efficiently perform a non-linear multivariate regression of local transport fluxes as a function of local dimensionless parameters. Transport predictions for thousands of DIII-D discharges showed excellent agreement with the power balance calculations across the whole plasma radius and over a broad range of operating regimes. Concerning predictive transport simulations, the framework made possible the design and automation of a workflow that enables self-consistent predictions of kinetic profiles and the plasma equilibrium. It is found that the feedback between the transport fluxes and plasma equilibrium can significantly affect the kinetic profiles predictions. Finally, such a rich set of results provide tangible evidence of how bottom-up approaches can potentially provide a fast track to integrated modeling solutions that are functional, cost-effective, and in sync with the research effort of the community.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗