Search NASA⌕ Search

SEARCH · Search NASA

Results for “Accelerators, Computing”

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 1,297 records · Page 72

TTDFT: A GPU accelerated Tucker tensor DFT code for large-scale Kohn-Sham DFT calculations

We present the Tucker tensor DFT (TTDFT) code which uses a tensor-structured algorithm with graphic processing unit (GPU) acceleration for conducting ground-state DFT calculations on large-scale systems. The Tucker tensor DFT algorithm uses a localized Tucker tensor basis computed from an additive separable approximation to the Kohn-Sham Hamiltonian. The discrete Kohn-Sham problem is solved using Chebyshev filtered subspace iteration method that relies on matrix-matrix multiplications of a sparse symmetric Hamiltonian matrix and a dense wavefunction matrix, expressed in the localized Tucker tensor basis. These matrix-matrix multiplication operations, which constitute the most computationally intensive step of the solution procedure, are GPU accelerated providing ~8-fold GPU-CPU speedup for these operations on the largest systems studied. In conclusion, the computational performance of the TTDFT code is presented using benchmark studies on aluminum nano-particles and silicon quantum dots with system sizes ranging up to ~7,000 atoms.

97 MATHEMATICS AND COMPUTING↗

FPGAs as a Service to Accelerate Machine Learning Inference [PowerPoint]

Large-scale particle physics experiments face challenging demands for high-throughput computing resources both now and in the future. New heterogeneous computing paradigms on dedicated hardware with increased parallelization, such as Field Programmable Gate Arrays (FPGAs), offer exciting solutions with large potential gains. The growing applications of machine learning algorithms in particle physics for simulation, reconstruction, and analysis are naturally deployed on such platforms. We demonstrate that the acceleration of machine learning inference as a web service represents a heterogeneous computing solution for particle physics experiments that requires minimal modification to the current computing model. As examples, we retrain the ResNet-50 convolutional neural network to demonstrate state-of-the-art performance for top quark jet tagging at the LHC, and apply a ResNet-50 model with transfer learning for neutrino event classification. Using Project Brainwave b y Microsoft to accelerate the ResNet-50 image classification model, we achieve average inference times of 60 (10) milliseconds with our experimental physics software framework using Brainwave as a cloud (edge or on-premises) service, representing an improvement by a factor of approximately 30 (175) in model inference latency over traditional CPU inference in current experimental hardware. A single FPGA service accessed by many CPUs achieves a throughput of 600--700 inferences per second using an image batch of one, comparable to large batch-size GPU throughput and significantly better than small batch-size GPU throughput. Deployed as an edge or cloud service for the particle physics computing model, coprocessor accelerators can have a higher duty cycle and are potentially much more cost-effective.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

MC21 analysis of the MIT PWR benchmark: Hot zero power results

MC21 Monte Carlo results have been compared with hot zero power measurements from an operating pressurized water reactor (PWR), as specified in a new full core PWR performance benchmark from the MIT Computational Reactor Physics Group. Included in the comparisons are axially integrated full core detector measurements, axial detector profiles, control rod bank worths, and temperature coefficients. Power depressions from grid spacers are seen clearly in the MC21 results. Application of Coarse Mesh Finite Difference (CMFD) acceleration within MC21 has been accomplished, resulting in a significant reduction of inactive batches necessary to converge the fission source. CMFD acceleration has also been shown to work seamlessly with the Uniform Fission Site (UFS) variance reduction method. (authors)

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

AMReX: Block-structured adaptive mesh refinement for multiphysics applications

Block-structured adaptive mesh refinement (AMR) provides the basis for the temporal and spatial discretization strategy for a number of Exascale Computing Project applications in the areas of accelerator design, additive manufacturing, astrophysics, combustion, cosmology, multiphase flow, and wind plant modeling. AMReX is a software framework that provides a unified infrastructure with the functionality needed for these and other AMR applications to be able to effectively and efficiently utilize machines from laptops to exascale architectures. AMR reduces the computational cost and memory footprint compared to a uniform mesh while preserving accurate descriptions of different physical processes in complex multiphysics algorithms. AMReX supports algorithms that solve systems of partial differential equations in simple or complex geometries and those that use particles and/or particle–mesh operations to represent component physical processes. In this article, we will discuss the core elements of the AMReX framework such as data containers and iterators as well as several specialized operations to meet the needs of the application projects. In addition, we will highlight the strategy that the AMReX team is pursuing to achieve highly performant code across a range of accelerator-based architectures for a variety of different applications.

Zhang, Weiqun↗

Improved parallel solution techniques for the integral transport matrix method

Alternative solution strategies to the parallel block Jacobi (PBJ) method for the solution of the global problem with the integral transport matrix method operators have been designed and tested. The most straightforward improvement to the Jacobi iterative method is the Gauss-Seidel alternative. The parallel red-black Gauss-Seidel (PGS) algorithm can improve on the number of iterations and reduce work per iteration by applying an alternating red-black color-set to the subdomains and assigning multiple sub-domains per processor. A parallel GMRES(m) method was implemented as an alternative to stationary iterations. Computational results show that the PGS method can improve on the PBJ method execution by up to {approx}50% when eight sub-domains per processor are used. However, compared to traditional source iterations with diffusion synthetic acceleration, it is still approximately an order of magnitude slower. The best-performing case are opticaUy thick because sub-domains decouple, yielding faster convergence. Further tests revealed that 64 sub-domains per processor was the best performing level of sub-domain division. An acceleration technique that improves the convergence rate would greatly improve the ITMM. The GMRES(m) method with a diagonal block preconditioner consumes approximately the same time as the PBJ solver but could be improved by an as yet undeveloped, more efficient preconditioner.

97 MATHEMATICS AND COMPUTING↗

Extending the distributed computing infrastructure of the CMS experiment with HPC resources

Particle accelerators are an important tool to study the fundamental properties of elementary particles. Currently the highest energy accelerator is the LHC at CERN, in Geneva, Switzerland. Each of its four major detectors, such as the CMS detector, produces dozens of Petabytes of data per year to be analyzed by a large international collaboration. The processing is carried out on the Worldwide LHC Computing Grid, that spans over more than 170 compute centers around the world and is used by a number of particle physics experiments. Recently the LHC experiments were encouraged to make increasing use of HPC resources. While Grid resources are homogeneous with respect to the used Grid middleware, HPC installations can be very different in their setup. In order to integrate HPC resources into the highly automatized processing setups of the CMS experiment a number of challenges need to be addressed. For processing, access to primary data and metadata as well as access to the software is required. At Grid sites all this is achieved via a number of services that are provided by each center. However at HPC sites many of these capabilities cannot be easily provided and have to be enabled in the user space or enabled by other means. At HPC centers there are often restrictions regarding network access to remote services, which is again a severe limitation. The paper discusses a number of solutions and recent experiences by the CMS experiment to include HPC resources in processing campaigns.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Efficient Parallelization of Irregular Applications on GPU Architectures

With the enlarging computation capacity of general Graphics Processing Units (GPUs), leveraging GPUs to accelerate parallel applications has become a critical topic in academia and industry. However, a wide range of irregular applications with the computation-/memory-intensive nature cannot easily achieve high GPU utilization. The challenges mainly involve the following aspects: first, data dependence leads to coarse-grained kernel and inefficient parallelism; second, heavy GPU memory usage may cause frequent memory evictions and extra overhead of I/O; third, specific computation patterns produce memory redundancies; last, workload balance and data reusability conjunctly benefit the overall performance, but there may exist a dynamic trade-off between them. Targeting these challenges, this dissertation proposes multiple optimizations to accelerate two real-world applications: many-body correlation functions to simulate nuclear physics in a large-scale scientific system; the other is the eALS-based matrix factorization recommendation system. To accelerate the calculations of many-body correlation functions, this dissertation presents three frameworks in GPU memory management and multi-GPU scheduling. Firstly, an optimized systematic GPU memory management framework, MemHC, utilizes a series of new memory reduction designs in GPU memory allocation, CPU/GPU communications, and GPU memory oversubscription. Secondly, an enhanced multi-GPU scheduling framework, MICCO, particularly by taking both data dimension (e.g., data reuse and data eviction) and computation dimension into account. MICCO designs a heuristic scheduling algorithm and a machine learning-based regression model to generate the optimal settings of a proposed new concept to manage the trade-off. Thirdly, a locality-aware multi-GPU scheduling framework. This scheduler leverages pipeline batch generation with a looking-ahead strategy by building local dependency graphs for memory transfer reduction and better data reuse, achieving up to 79.92% memory cost reduction and 1.67x speedup. To parallelize the eALS-based recommendation system, this dissertation proposes an efficient CPU/GPU heterogeneous recommendation system, HEALS. HEALS employs newly designed architecture-adaptive data formats to achieve load balance and good data locality on CPU and GPU. To mitigate the data dependence, HEALS presents a CPU/GPU collaboration model for both task parallelism and data parallelism with multiple kernel computation optimizations. In summary, this dissertation efficiently accelerates two typical irregular applications on GPUs by building four frameworks, including CPU/GPU collaboration, GPU memory management, and multi-GPU scheduling.

Wang, Qihan↗

Generic Multi-Layer Perceptron Inference Accelerator on FPGA (vneuron) v1.0

We have designed and implemented a neural network inference compute engine (vneuron) that can be deployed in the fabric of any FPGA without using special hardware accelerator primitive. The "vneuron" is purely written in verilog, and supports scalable neural network structure with fully connected layers and ReLU activation ( Multi-Layer Perceptron architecture) with 16 bits of precision. We have demonstrated it on an Xilinx Artix 7 FPGA for a 16-input, 8-output MLP with 3 layer, 1600 parameters. It takes 40 DSP48E and 40 BRAM18, and takes 131 clock cycles for computing (1048 ns when clocked at 125MHz). We include PyTorch quantization from a given floating point model, and provide behavioral verification simulation in the disclosed software package.

Du, Qiang↗

WarpX: Delivering Unprecedented Particle-in-Cell Simulation Capability

Since 1939, more than 30 percent of all Nobel prizes in physics and four of the past 14 prizes in chemistry have been awarded for work with particle accelerators. These tools are now used to treat cancer and produce medical supplies, support research in applied fields such as pharmaceuticals, create microcircuits, and sterilize food—and the applications list continues to grow. However, the size of the facilities needed to house these machines—which are often dozens of kilometers long—has limited their future potential. Thus, researchers are motivated to create novel particle accelerator designs that provide new imaging capability but have a much smaller footprint. The Exascale Computing Project’s WarpX application captures the full complexity of the acceleration processes within plasma-based design, supporting the rapid and cost-efficient development of tens of thousands of new particle accelerators for various applications from improving human health to nuclear fusion reactors.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Virtual Observation System for Earth System Model: An Application to ACME Land Model Simulations

Investigating and evaluating physical-chemical-biological processes within an Earth system model (EMS) can be very challenging due to the complexity of both model design and software implementation. A virtual observation system (VOS) is presented to enable interactive observation of these processes during system simulation. Based on advance computing technologies, such as compiler-based software analysis, automatic code instrumentation, and high-performance data transport, the VOS provides run-time observation capability, in-situ data analytics for Earth system model simulation, model behavior adjustment opportunities through simulation steering. A VOS for a terrestrial land model simulation within the Accelerated Climate Modeling for Energy model is also presented to demonstrate the implementation details and system innovations.

58 GEOSCIENCES↗

Speeding up equation of motion coupled cluster theory with the chain of spheres approximation

In the present paper, the chain of spheres exchange (COSX) approximation is applied to the highest scaling terms in the equation of motion (EOM) coupled cluster equations with single and double excitations, in particular, the terms involving integrals with four virtual labels. It is found that even the acceleration of this single term yields significant computational gains without compromising the desired accuracy of the method. For an excitation energy calculation on a cluster of five water molecules using 585 basis functions, the four virtual term is 9.4 times faster using COSX with a loose grid than using the canonical implementation, which yields a 2.6 fold acceleration for the whole of the EOM calculation. For electron attachment calculations, the four virtual term is 15 times and the total EOM calculation is 10 times faster than the canonical calculation for the same system. The accuracy of the new method was tested using Thiel’s test set for excited states using the same settings and the maximum absolute deviation over the whole test set was found to be 12.945 cm{sup −1} (59 μHartree) for excitation energies and 6.799 cm{sup −1} (31 μHartree) for electron attachments. Using MP2 amplitudes for the ground state in combination with the parallel evaluation of the full EOM equations in the manner discussed in this paper enabled us to perform calculations for large systems. Electron affinity values for the two lowest states of a Zn protoporphyrine model compound (224 correlated electrons and 1120 basis functions) were obtained in 3 days 19 h using 4 cores of a Xeon E5-2670 processor allocating 10 GB memory per core. Calculating the lowest two excitation energies for trans-retinal (114 correlated electrons and 539 basis functions) took 1 day 21 h using eight cores of the same processor and identical memory allocation per core.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

An Easy Method to Accelerate an Iterative Algebraic Solver, part II

This study is a companion to [Yao, Journal of Computational Physics267(2014)]. We provide a compact alternative development of Yao’s method to accelerate to order 2n–1 a fixed-point iterative scheme with an original convergence-rate of ordern. Using this approach, we show how to further improve the order of convergence by increasings, the number of steps per iteration. This scheme extends to arbitrarily high order without extra derivative evaluations per iteration. We discuss the efficiency of the new methods, including s = 2 (Yao’s method), and compare to that of the original method as a function of n; and we consider the efficiency for systems of equations of system-size M. For n = 2 and large M, we find that the multi-step scheme peaks in efficiency at about s ln s ~ M function evaluations, where its computational speed is several times faster than Newton’s method.

97 MATHEMATICS AND COMPUTING↗

AXI4MLIR: User-Driven Automatic Host Code Generation for Custom AXI-Based Accelerators

Tensor algebra operations represent an important class of algorithms used across many applications, including machine learning, scientific computing, and data analytics. As a result, the efficient generation of custom accelerators for tensor operations has received increased attention. Previous efforts have produced automated tools enabling users to prototype and explore optimized accelerators. However, little effort has been focused on the host-accelerator interaction in these tools. Efficient use of hardware accelerators requires knowledge about the accelerator's capabilities (operations, data formats, and opcode support), the host CPU microarchitecture (e.g., memory hierarchy), the host-accelerator interface, and the application's features (which code regions should be mapped onto an accelerator). Manually rewriting the original applications to facilitate improved custom accelerator mapping is an error-prone and time-consuming endeavor. To cope with this, we propose AXI4MLIR, a new framework to automatically generate and optimize the communication between the host CPU and arbitrary accelerators that implement linear algebra algorithms. AXI4MLIR extends the MLIR compiler framework to automatically generate efficient host-accelerator driver code for accelerators with AXI-based interfaces. Our compiler extensions enable automatic driver code generation while carefully considering the host's memory hierarchy and target accelerator features. To demonstrate the flexibility and utility of AXI4MLIR, we test it with diverse use cases that include different types of accelerators, tiling scenarios, and dataflow schemes. We compare our experimental results to manual implementations of host-accelerator driver code and find that our approach can reduce CPU cache references by 56% and deliver up to a 1.65x speedup.

Bohm Agostini, Nicolas↗

Modeling Codes Used for LAMP

LAMP team is planning on using multiple computer codes as tools for modeling beam dynamics, electromagnetics, and plasma environment in ion sources. Below is a description of the computer tools we have been using or planning to use in the future. Most of the codes chosen for the LAMP project are capable to model most of the beam dynamics in LAMP. A judicious choice of the appropriate code to use will be based on speed and fidelity of the modeling with faster calculations produced by simplifying some features.

43 PARTICLE ACCELERATORS↗

On the integration of molecular dynamics, data science, and experiments for studying solvent effects on catalysis

Computational workflows that combine molecular dynamics (MD) simulations and emerging data-centric (DC) methods can accelerate the screening and analysis of solvent systems experimentally and computationally. Here, MD simulations provide atomic positions and velocities of reactant, solvent, and catalyst materials that can be manipulated into data representations that in turn can be used by DC techniques to conduct predictive modeling, feature extraction, and experimental design. For liquid-phase catalytic applications, emerging DC techniques such as Convolutional and Graph Neural Networks (CNN/GNN), Topological Data Analysis (TDA), and Active Learning (AL) can leverage MD and experimental data to quickly predict solvent effects on reaction outcomes. For instance, in recent studies, 3D solvent environments obtained with MD have been exploited by CNNs to predict experimental reaction rates for homogeneous acid-catalyzed lignocellulosic processes. In this perspective, we discuss basic principles of DC methods and how these can be combined with MD to enable high-throughput screening of solvent selection for diverse catalysis applications.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Benchmark studies among own and commercial beam tracking codes on fundamental conditions

In order to design accelerators or to analyze experimental data, commercially or noncommercially programs for tracking particles have been developed and used. However the reliabilities of those programs are not clear, especially if the beam quality is as high as that required for future applications. Therefore, for the purpose to study the reliabilities and to improve their accuracy, we planed to benchmark those programs. As a first step, the calculations were carried out in the both condition of free space and homogeneously accelerating-field by using three own codes and three commercial codes (MAGIC, PARMELA, and GPT)

43 PARTICLE ACCELERATORS↗