Search NASA⌕ Search

SEARCH · Search NASA

Results for “floating-point data”

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.

71 records · Page 4

Quantifying the Impact of Single Bit Flips on Floating Point Arithmetic

In high-end computing, the collective surface area, smaller fabrication sizes, and increasing density of components have led to an increase in the number of observed bit flips. If mechanisms are not in place to detect them, such flips produce silent errors, i.e. the code returns a result that deviates from the desired solution by more than the allowed tolerance and the discrepancy cannot be distinguished from the standard numerical error associated with the algorithm. These phenomena are believed to occur more frequently in DRAM, but logic gates, arithmetic units, and other circuits are also susceptible to bit flips. Previous work has focused on algorithmic techniques for detecting and correcting bit flips in specific data structures, however, they suffer from lack of generality and often times cannot be implemented in heterogeneous computing environment. Our work takes a novel approach to this problem. We focus on quantifying the impact of a single bit flip on specific floating-point operations. We analyze the error induced by flipping specific bits in the most widely used IEEE floating-point representation in an architecture-agnostic manner, i.e., without requiring proprietary information such as bit flip rates and the vendor-specific circuit designs. We initially study dot products of vectors and demonstrate that not all bit flips create a large error and, more importantly, expected value of the relative magnitude of the error is very sensitive on the bit pattern of the binary representation of the exponent, which strongly depends on scaling. Our results are derived analytically and then verified experimentally with Monte Carlo sampling of random vectors. Furthermore, we consider the natural resilience properties of solvers based on the fixed point iteration and we demonstrate how the resilience of the Jacobi method for linear equations can be significantly improved by rescaling the associated matrix.

Elliott, James J.↗

Exploiting data representation for fault tolerance

Incorrect computer hardware behavior may corrupt intermediate computations in numerical algorithms, possibly resulting in incorrect answers. Prior work models misbehaving hardware by randomly flipping bits in memory. We start by accepting this premise, and present an analytic model for the error introduced by a bit flip in an IEEE 754 floating-point number. We then relate this finding to the linear algebra concepts of normalization and matrix equilibration. In particular, we present a case study illustrating that normalizing both vector inputs of a dot product minimizes the probability of a single bit flip causing a large error in the dot product's result. Moreover, the absolute error is either less than one or very large, which allows detection of large errors. Then, we apply this to the GMRES iterative solver. We count all possible errors that can be introduced through faults in arithmetic in the computationally intensive orthogonalization phase of GMRES, and show that when the matrix is equilibrated, the absolute error is bounded above by one.

97 MATHEMATICS AND COMPUTING↗

Leveraging In-Network Computing and Programmable Switches for Streaming Analysis of Scientific Data

With the emergence of programmable network devices that match the performance of fixed function devices, several recent projects have explored in-network computing, where the processing that is traditionally done outside the network is offloaded to the network devices. In-network computing has typically been applied to network functions (e.g., load balancing, NAT, and DNS), caching, data reduction/aggregation, and coordination/consensus functions. In some cases it has been used to accelerate stream-processing tasks that involve small payloads and simple operations. In this work we focus on leveraging in-network computing for stream processing of scientific datasets with large payloads that require complex operations such as floating-point computations and logarithmic functions. We demonstrate in-network computing for a real-world scientific application performing streaming normalization of a 2-D image from a light source experiment. We discuss the challenges we encountered and potential approaches to address them.

Sankaran, Ganesh↗

Sparse triangular solves for ILU revisited: data layout crucial to better performance.; .

A key to good processor utilization for sparse matrix computations is storing the data in the format that is most conducive to fast access by the memory system. In particular, for sparse matrix triangular solves the traditional compressed sparse matrix format is poor, and minor adjustments to the data structure can increase the processor utilization dramatically. Such adjustments involve storing the L and U factors separately and storing the U rows 'backwards' so that they are accessed in a simple streaming fashion during the triangular solves. Changes to the PETSc libraries to use this modified storage format resulted in over twice the floating-point rate for some matrices. This improvement can be accounted for by a decrease in the cache misses and TLB (transaction lookaside buffer) misses in the modified code.

99 GENERAL AND MISCELLANEOUS↗

Leveraging explainable AI to characterize floating-point exceptions in linear solvers

Linear solver packages are central to many scientific, engineering, and machine learning applications. When floating-point exceptions occur in these solvers, e.g., division by zero or overflow, numerical results are compromised and become unreliable. Existing static and dynamic analysis tools can detect such exceptions, but they do not explain why the exceptions occur in terms of the solver inputs. Here, we present a study to characterize the inputs that cause numerical exceptions in linear solver packages. Our approach uses explainable AI (XAI) to find the most relevant characteristics of input matrices that explain the occurrence of exceptions in the solvers. Since training data in this domain is scarce, we perform extensive data gathering and data augmentation to obtain exception-inducing inputs. Our approach uses a repair strategy on the features blamed by XAI to validate that such features indeed explain the exceptions. We compare the LIME and SHAP XAI techniques using a dozen matrix features with three classifiers. We evaluate the approach on three widely used linear solver packages and find that some input characteristics can explain the occurrence of exceptions 100% of the time, in specific solvers and preconditioners.

Explainable AI↗

Empirically-calibrated H100 node power models for accurate AI training energy estimation

Accurately quantifying the energy use of artificial intelligence (AI) training is critical for infrastructure planning, carbon accounting, and sustainable data center operation, but few studies have directly measured the power consumption of production workloads on contemporary hardware. By combining empirical measurements from Brookhaven National Laboratory during AI training on 8-graphics-processing-unit H100 systems with open-source benchmarking data, we develop statistical models relating computational intensity to node-level power consumption. We measure the gap between manufacturer-rated thermal design power (TDP) and actual power demand during AI training. Our analysis reveals that even computationally intensive workloads operate at only 76% of the 10.2 kW TDP rating. Our architecture-specific model, calibrated to floating-point operations, predicts energy consumption with 11.4% mean absolute percentage error, significantly outperforming TDP-based approaches (27%–37% error). We identified distinct power signatures between transformer and convolutional neural network architectures, with transformers showing characteristic fluctuations that may impact grid stability. These results provide a measurement-grounded basis for improving AI training energy estimates, enabling more reliable infrastructure sizing, cost projections, and environmental impact assessments.

Newkirk, Alex C↗

An efficient mixed-precision, hybrid CPU-GPU implementation of a nonlinearly implicit one-dimensional particle-in-cell algorithm

Recently, a fully implicit, energy- and charge-conserving particle-in-cell method has been developed for multi-scale, full-f kinetic simulations [G. Chen, et al., J. Comput. Phys. 230, 18 (2011)]. The method employs a Jacobian-free Newton-Krylov (JFNK) solver and is capable of using very large timesteps without loss of numerical stability or accuracy. A fundamental feature of the method is the segregation of particle orbit integrations from the field solver, while remaining fully self-consistent. This provides great flexibility, and dramatically improves the solver efficiency by reducing the degrees of freedom of the associated nonlinear system. However, it requires a particle push per nonlinear residual evaluation, which makes the particle push the most time-consuming operation in the algorithm. This paper describes a very efficient mixed-precision, hybrid CPU-GPU implementation of the implicit PIC algorithm. The JFNK solver is kept on the CPU (in double precision), while the inherent data parallelism of the particle mover is exploited by implementing it in single-precision on a graphics processing unit (GPU) using CUDA. Performance-oriented optimizations, with the aid of an analytical performance model, the roofline model, are employed. Despite being highly dynamic, the adaptive, charge-conserving particle mover algorithm achieves up to 300 400 GOp/s (including single-precision floating-point, integer, and logic operations) on a Nvidia GeForce GTX580, corresponding to 20 25% absolute GPU efficiency (against the peak theoretical performance) and 50-70% intrinsic efficiency (against the algorithm s maximum operational throughput, which neglects all latencies). This is about 200-300 times faster than an equivalent serial CPU implementation. When the single-precision GPU particle mover is combined with a double-precision CPU JFNK field solver, overall performance gains 100 vs. the double-precision CPU-only serial version are obtained, with no apparent loss of robustness or accuracy when applied to a challenging long-time scale ion acoustic wave simulation.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Livermore Compiler Analysis Loop Suite

LCALS is designed to evaluate compiler optimizations and performance of a variety of loop kernels and loop traversal software constructs. Some of the loop kernels are pulled directly from "Livermore Loops Coded in C", developed at LLNL (see item 11 below for details of earlier code versions). The older suites were used to evaluate floating-point performances of hardware platforms prior to porting larger application codes. The LCALS suite is geared toward assissing C++ compiler optimizations and platform performance related to SIMD vectorization, OpenMP threading, and advanced C++ language features. LCALS contains 20 of 24 loop kernels from the older Livermore Loop suites, plus various others representative of loops found in current production appkication codes at LLNL. The latter loops emphasize more diverse loop constructs and data access patterns than the others, such as multi-dimensional difference stencils. The loops are included in a configurable framework, which allows control of compilation, loop sampling for execution timing, which loops are run and their lengths. It generates timing statistics for analysis and comparing variants of individual loops. Also, it is easy to add loops to the suite as desired.

Hornung, R. D.↗

GeantV alpha release

In the fall 2016, GeantV went through a thorough community evaluation of the project status and of its strategy for sharing the R&D results with the LHC experiments and with the HEP simulation community in general. Following this discussion, GeantV has engaged onto an ambitious 2-year road-path aiming to deliver a beta version that has most of the final design and several performance features of the final product, partially integrated with some of the experiment’s frameworks. The initial GeantV prototype has been updated to a vector-aware concurrent framework, which is able to deliver high-density floating-point computations for most of the performance-critical components such as propagation in field and physics models. Electromagnetic physics models were adapted for the specific GeantV requirements, aiming for the full demonstration of shower physics performance in the alpha release at the end of 2017. We have revisited and formalized GeantV user interfaces and helper protocols, allowing to: connect to user code, provide recipes to access efficiently MC truth and generate user data in a concurrent environment.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

ECP Milestone Memo WBS 2.3.4.13 ECP/VTK-m FY19Q1 [MS-19/01-03] ZFP / Release / Clip STDA05-17

The STDA05-17 milestone comprises the following 3 deliverables. VTK-m Release 2 We will provide a release of VTK-m software and associated documentation. The source code repository will be tagged at a stable state, and, at a minimum, tarball captures of the source code will be made available from the web site. A version of the VTK-m User's Guide documenting this release will also be made available. Productionize zfp compression The "ZFP: Compressed Floating-Point Arrays" project (WBS 1.3.4.13) is creating an implementation of ZFP compression in VTK-m. Their implementation will be focused on operating in CUDA. The VTK-m project will assist by generalizing the implementation to other devices (such as multi-core CPUs). We will also assist in productionizing the code such that it can be used by external projects and products. Clip Clip operations intersect meshes with implicit functions. It is the foundation of spatial subsetting algorithms, such as "box," and the foundation of data-based subsetting, such as "isovolume." The algorithm requires considering thousands of possible cases, and is thus quite difficult to implement. This milestone will implement clipping to be sufficient for Visit's and ParaView's needs.

97 MATHEMATICS AND COMPUTING↗

Resiliency in numerical algorithm design for extreme scale simulations

Here this work is based on the seminar titled ‘Resiliency in Numerical Algorithm Design for Extreme Scale Simulations’ held March 1–6, 2020, at Schloss Dagstuhl, that was attended by all the authors. Advanced supercomputing is characterized by very high computation speeds at the cost of involving an enormous amount of resources and costs. A typical large-scale computation running for 48 h on a system consuming 20 MW, as predicted for exascale systems, would consume a million kWh, corresponding to about 100k Euro in energy cost for executing 10 23 floating-point operations. It is clearly unacceptable to lose the whole computation if any of the several million parallel processes fails during the execution. Moreover, if a single operation suffers from a bit-flip error, should the whole computation be declared invalid? What about the notion of reproducibility itself: should this core paradigm of science be revised and refined for results that are obtained by large-scale simulation? Naive versions of conventional resilience techniques will not scale to the exascale regime: with a main memory footprint of tens of Petabytes, synchronously writing checkpoint data all the way to background storage at frequent intervals will create intolerable overheads in runtime and energy consumption. Forecasts show that the mean time between failures could be lower than the time to recover from such a checkpoint, so that large calculations at scale might not make any progress if robust alternatives are not investigated. More advanced resilience techniques must be devised. The key may lie in exploiting both advanced system features as well as specific application knowledge. Research will face two essential questions: (1) what are the reliability requirements for a particular computation and (2) how do we best design the algorithms and software to meet these requirements? While the analysis of use cases can help understand the particular reliability requirements, the construction of remedies is currently wide open. One avenue would be to refine and improve on system- or application-level checkpointing and rollback strategies in the case an error is detected. Developers might use fault notification interfaces and flexible runtime systems to respond to node failures in an application-dependent fashion. Novel numerical algorithms or more stochastic computational approaches may be required to meet accuracy requirements in the face of undetectable soft errors. These ideas constituted an essential topic of the seminar. The goal of this Dagstuhl Seminar was to bring together a diverse group of scientists with expertise in exascale computing to discuss novel ways to make applications resilient against detected and undetected faults. In particular, participants explored the role that algorithms and applications play in the holistic approach needed to tackle this challenge. This article gathers a broad range of perspectives on the role of algorithms, applications and systems in achieving resilience for extreme scale simulations. The ultimate goal is to spark novel ideas and encourage the development of concrete solutions for achieving such resilience holistically.

79 ASTRONOMY AND ASTROPHYSICS↗

Fluid-Kinetic Coupling: Advanced Discretizations for Simulations on Emerging Heterogeneous Architectures (LDRD FY20-0643)

Plasma physics simulations are vital for a host of Sandia mission concerns, for fundamental science, and for clean energy in the form of fusion power. Sandia's most mature plasma physics simulation capabilities come in the form of particle-in-cell (PIC) models and magnetohydrodynamics (MHD) models. MHD models for a plasma work well in denser plasma regimes when there is enough material that the plasma approximates a fluid. PIC models, on the other hand, work well in lower-density regimes, in which there is not too much to simulate; error in PIC scales as the square root of the number of particles, making high-accuracy simulations expensive. Real-world applications, however, almost always involve a transition region between the high-density regimes where MHD is appropriate, and the low-density regimes for PIC. In such a transition region, a direct discretization of Vlasov is appropriate. Such discretizations come with their own computational costs, however; the phase-space mesh for Vlasov can involve up to six dimensions (seven if time is included), and to apply appropriate homogeneous boundary conditions in velocity space requires meshing a substantial padding region to ensure that the distribution remains sufficiently close to zero at the velocity boundaries. Moreover, for collisional plasmas, the right-hand side of the Vlasov equation is a collision operator, which is non-local in velocity space, and which may dominate the cost of the Vlasov solver. The present LDRD project endeavors to develop modern, foundational tools for the development of continuum-kinetic Vlasov solvers, using the discontinuous Petrov-Galerkin (DPG) methodology, for discretization of Vlasov, and machine-learning (ML) models to enable efficient evaluation of collision operators. DPG affords several key advantages. First, it has a built-in, robust error indicator, allowing us to adapt the mesh in a very natural way, enabling a coarse velocity-space mesh near the homogeneous boundaries, and a fine mesh where the solution has fine features. Second, it is an inherently high-order, high-intensity method, requiring extra local computations to determine so-called optimal test functions, which makes it particularly suited to modern hardware in which floating-point throughput is increasing at a faster rate than memory bandwidth. Finally, DPG is a residual-minimizing method, which enables high-accuracy computation: in typical cases, the method delivers something very close to the $L^2$ projection of the exact solution. Meanwhile, the ML-based collision model we adopt affords a cost structure that scales as the square root of a standard direct evaluation. Moreover, we design our model to conserve mass, momentum, and energy by construction, and our approach to training is highly flexible, in that it can incorporate not only synthetic data from direct-simulation Monte Carlo (DSMC) codes, but also experimental data. We have developed two DPG formulations for Vlasov-Poisson: a time-marching, backward-Euler discretization and a space-time discretization. We have conducted a number of numerical experiments to verify the approach in a 1D1V setting. In this report, we detail these formulations and experiments. We also summarize some new theoretical results developed as part of this project (published as papers previously): some new analysis of DPG for the convection-reaction problem (of which the Vlasov equation is an instance), a new exponential integrator for DPG, and some numerical exploration of various DPG-based time-marching approaches to the heat equation. As part of this work, we have contributed extensively to the Camellia open-source library; we also describe the new capabilities and their usage. We have also developed a well-documented methodology for single-species collision operators, which we applied to argon and demonstrated with numerical experiments. We summarize those results here, as well as describing at a high level a design extending the methodology to multi-species operators. We have released a new open-source library, MLC, under a BSD license; we include a summary of its capabilities as well.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

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↗

Experimental and numerical investigations of a two-body floating-point absorber wave energy converter in regular waves

This paper presents experimental and numerical studies on the hydrodynamics of a two-body floating-point absorber (FPA) wave energy converter (WEC) under both extreme and operational wave conditions. In this study, the responses of the WEC in heave, surge, and pitch were evaluated for various regular wave conditions. For extreme condition analysis, we assume the FPA system has a survival mode that locks the power-take-off (PTO) mechanism in extreme waves, and the WEC moves as a single body in this scenario. A series of Reynolds-averaged Navier-Stokes (RANS) simulations was performed for the survival condition analysis, and the results were validated with the measurements from experimental wave tank tests. For the FPA system in operational conditions, both a boundary element method (BEM) and the RANS simulations were used to analyze the motion response and power absorption performance. Additional viscous damping, primarily induced by flow separation and vortex shedding, is included in the BEM simulation as a quadratic drag force proportional to the square of body velocity. The inclusion of viscous drag improves the accuracy of BEM's prediction of the heave response and the power absorption performance of the FPA system, which agrees well with experimental data and the RANS simulation results over a broad range of incident wave periods, except near resonance in large wave height scenarios. Overall, the experimental and numerical results suggest that nonlinear effects, caused by viscous damping and interaction between waves and the FPA, significantly influence the system response and power absorption performance. Nonlinear effects were found to be particularly significant when the wave height was large and the period was near or shorter than the resonant period of the FPA system. The study is expected to be helpful for understanding the nonlinear interaction between waves and the FPA system so that the structure of the FPA can be adequately designed.

16 TIDAL AND WAVE POWER↗

Scalable Knowledge Graph Analytics at 136 Petaflop/s

We are motivated by newly proposed methods for data mining large-scale corpora of scholarly publications, such as the full biomedical literature, which may consist of tens of millions of papers spanning decades of research. In this setting, analysts seek to discover how concepts relate to one another. They construct graph representations from annotated text databases and then formulate the relationship-mining problem as one of computing all-pairs shortest paths (APSP), which becomes a significant bottleneck. In this context, we present a new high-performance algorithm and implementation of the Floyd-Warshall algorithm for distributed-memory parallel computers accelerated by GPUs, which we call DSNAPSHOT (Distributed Accelerated Semiring All-Pairs Shortest Path). For our largest experiments, we ran DSNAPSHOT on a connected input graph with millions of vertices using 4, 096nodes (24,576GPUs) of the Oak Ridge National Laboratory's Summit supercomputer system. We find DSNAPSHOT achieves a sustained performance of 136×1015 floating-point operations per second (136petaflop/s) at a parallel efficiency of 90% under weak scaling and, in absolute speed, 70% of the best possible performance given our computation (in the single-precision tropical semiring or “min-plus” algebra). Looking forward, we believe this novel capability will enable the mining of scholarly knowledge corpora when embedded and integrated into artificial intelligence-driven natural language processing workflows at scale.

Kannan, Ramakrishnan {ramki}↗

Fast and Accurate Intersections on a Sphere

We introduce a fast, high-precision algorithm for calculating intersections between great circle arcs and lines of constant latitude on the unit sphere. We first propose a simplified intersection point formula with improved speed and numerical robustness over the ones traditionally implemented in geoscience software. We then show how algorithms based on the concept of error-free transformations (EFT) can be applied to evaluate this formula within a relative error bound that is on the order of machine precision. Here, we demonstrate that, with a vectorized and parallelized implementation, this enhanced accuracy is achieved with no compute time overhead compared to a direct calculation in hardware floating point, making our algorithm suitable for performance-sensitive applications like regridding of high-resolution climate data. In contrast, evaluating our formula using high-precision data types like quadruple precision and arbitrary precision, or using the robust intersection computation routines from the Computational Geometry Algorithms Library, leads to significant computational overhead, especially since these alternatives inhibit vectorization. More generally, our work demonstrates how EFT techniques can be combined and extended to implement nontrivial geometric calculations with high accuracy and speed.

Environmental sciences↗

Implementing Molecular Dynamics on Hybrid High Performance Computers - Particle-Particle Particle-Mesh

The use of accelerators such as graphics processing units (GPUs) has become popular in scientific computing applications due to their low cost, impressive floating-point capabilities, high memory bandwidth, and low electrical power requirements. Hybrid high-performance computers, machines with nodes containing more than one type of floating-point processor (e.g. CPU and GPU), are now becoming more prevalent due to these advantages. In this paper, we present a continuation of previous work implementing algorithms for using accelerators into the LAMMPS molecular dynamics software for distributed memory parallel hybrid machines. In our previous work, we focused on acceleration for short-range models with an approach intended to harness the processing power of both the accelerator and (multi-core) CPUs. To augment the existing implementations, we present an efficient implementation of long-range electrostatic force calculation for molecular dynamics. Specifically, we present an implementation of the particle-particle particle-mesh method based on the work by Harvey and De Fabritiis. We present benchmark results on the Keeneland InfiniBand GPU cluster. We provide a performance comparison of the same kernels compiled with both CUDA and OpenCL. We discuss limitations to parallel efficiency and future directions for improving performance on hybrid or heterogeneous computers.

97 MATHEMATICS AND COMPUTING↗