Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “GPU acceleration”

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 937 records · Page 52

Efficient Mixed-Precision Matrix Factorization of the Inverse Overlap Matrix in Electronic Structure Calculations with AI-Hardware and GPUs

In recent years, a new kind of accelerated hardware has gained popularity in the artificial intelligence (AI) community which enables extremely high-performance tensor contractions in reduced precision for deep neural network calculations. In this article, we exploit Nvidia Tensor cores, a prototypical example of such AI-hardware, to develop a mixed precision approach for computing a dense matrix factorization of the inverse overlap matrix in electronic structure theory, S –1 . This factorization of S –1 , written as ZZT = S –1 , is used to transform the general matrix eigenvalue problem into a standard matrix eigenvalue problem. Here we present a mixed precision iterative refinement algorithm where Z is given recursively using matrix–matrix multiplications and can be computed with high performance on Tensor cores. To understand the performance and accuracy of Tensor cores, comparisons are made to GPU-only implementations in single and double precision. Additionally, we propose a nonparametric stopping criteria which is robust in the face of lower precision floating point operations. The algorithm is particularly useful when we have a good initial guess to Z, for example, from previous time steps in quantum-mechanical molecular dynamics simulations or from a previous iteration in a geometry optimization.

36 MATERIALS SCIENCE↗

INTEGRATE - Inverse Network Transformations for Efficient Generation of Robust Airfoil and Turbine Enhancements

The INTEGRATE (Inverse Network Transformations for Efficient Generation of Robust Airfoil and Turbine Enhancements) project is developing a new inverse-design capability for the aerodynamic design of wind turbine rotors using invertible neural networks. This AI-based design technology can capture complex non-linear aerodynamic effects while being 100 times faster than design approaches based on computational fluid dynamics. This project enables innovation in wind turbine design by accelerating time to market through higher-accuracy early design iterations to reduce the levelized cost of energy. INVERTIBLE NEURAL NETWORKS Researchers are leveraging a specialized invertible neural network (INN) architecture along with the novel dimension-reduction methods and airfoil/blade shape representations developed by collaborators at the National Institute of Standards and Technology (NIST) learns complex relationships between airfoil or blade shapes and their associated aerodynamic and structural properties. This INN architecture will accelerate designs by providing a cost-effective alternative to current industrial aerodynamic design processes, including: - Blade element momentum (BEM) theory models: limited effectiveness for design of offshore rotors with large, flexible blades where nonlinear aerodynamic effects dominate - Direct design using computational fluid dynamics (CFD): cost-prohibitive - Inverse-design models based on deep neural networks (DNNs): attractive alternative to CFD for 2D design problems, but quickly overwhelmed by the increased number of design variables in 3D problems AUTOMATED COMPUTATIONAL FLUID DYNAMICS FOR TRAINING DATA GENERATION - MERCURY FRAMEWORK The INN is trained on data obtained using the University of Marylands (UMD) Mercury Framework, which has with robust automated mesh generation capabilities and advanced turbulence and transition models validated for wind energy applications. Mercury is a multi-mesh paradigm, heterogeneous CPU-GPU framework. The framework incorporates three flow solvers at UMD, 1) OverTURNS, a structured solver on CPUs, 2) HAMSTR, a line based unstructured solver on CPUs, and 3) GARFIELD, a structured solver on GPUs. The framework is based on Python, that is often used to wrap C or Fortran codes for interoperability with other solvers. Communication between multiple solvers is accomplished with a Topology Independent Overset Grid Assembler (TIOGA). NOVEL AIRFOIL SHAPE REPRESENTATIONS USING GRASSMAN SPACES We developed a novel representation of shapes which decouples affine-style deformations from a rich set of data-driven deformations over a submanifold of the Grassmannian. The Grassmannian representation as an analytic generative model, informed by a database of physically relevant airfoils, offers (i) a rich set of novel 2D airfoil deformations not previously captured in the data , (ii) improved low-dimensional parameter domain for inferential statistics informing design/manufacturing, and (iii) consistent 3D blade representation and perturbation over a sequence of nominal shapes. TECHNOLOGY TRANSFER DEMONSTRATION - COUPLING WITH NREL WISDEM Researchers have integrated the inverse-design tool for 2D airfoils (INN-Airfoil) into WISDEM (Wind Plant Integrated Systems Design and Engineering Model), a multidisciplinary design and optimization framework for assessing the cost of energy, as part of tech-transfer demonstration. The integration of INN-Airfoil into WISDEM allows for the design of airfoils along with the blades that meet the dynamic design constraints on cost of energy, annual energy production, and the capital costs. Through preliminary studies, researchers have shown that the coupled INN-Airfoil + WISDEM approach reduces the cost of energy by around 1% compared to the conventional design approach. This page will serve as a place to easily access all the publications from this work and the repositories for the software developed and released through this pr...

aerodynamics↗

Enabling Parallel Performance and Portability of Solid Mechanics Simulations Across CPU and GPU Architectures

Efficiently simulating solid mechanics is vital across various engineering applications. As constitutive models grow more complex and simulations scale up in size, harnessing the capabilities of modern computer architectures has become essential for achieving timely results. This paper presents advancements in running parallel simulations of solid mechanics on multi-core CPUs and GPUs using a single-code implementation. This portability is made possible by the C++ matrix and array (MATAR) library, which interfaces with the C++ Kokkos library, enabling the selection of fine-grained parallelism backends (e.g., CUDA, HIP, OpenMP, pthreads, etc.) at compile time. MATAR simplifies the transition from Fortran to C++ and Kokkos, making it easier to modernize legacy solid mechanics codes. We applied this approach to modernize a suite of constitutive models and to demonstrate substantial performance improvements across different computer architectures. This paper includes comparative performance studies using multi-core CPUs along with AMD and NVIDIA GPUs. Results are presented using a hypoelastic–plastic model, a crystal plasticity model, and the viscoplastic self-consistent generalized material model (VPSC-GMM). The results underscore the potential of using the MATAR library and modern computer architectures to accelerate solid mechanics simulations.

Morgan, Nathaniel (ORCID:0000000276118449)↗

Quantum-Based Molecular Dynamics Simulations Using Tensor Cores

Tensor cores, along with tensor processing units, represent a new form of hardware acceleration specifically designed for deep neural network calculations in artificial intelligence applications. Tensor cores provide extraordinary computational speed and energy efficiency but with the caveat that they were designed for tensor contractions (matrix–matrix multiplications) using only low-precision floating-point operations. Despite this perceived limitation, we demonstrate how tensor cores can be applied with high efficiency to the challenging and numerically sensitive problem of quantum-based Born–Oppenheimer molecular dynamics, which requires highly accurate electronic structure optimizations and conservative force evaluations. The interatomic forces are calculated on-the-fly from an electronic structure that is obtained from a generalized deep neural network, where the computational structure naturally takes advantage of the exceptional processing power of the tensor cores and allows for high performance in excess of 100 Tflops on a single Nvidia A100 GPU. Stable molecular dynamics trajectories are generated using the framework of extended Lagrangian Born–Oppenheimer molecular dynamics, which combines computational efficiency with long-term stability, even when using approximate charge relaxations and force evaluations that are limited in accuracy by the numerically noisy conditions caused by the low-precision tensor core floating-point operations. A canonical ensemble simulation scheme is also presented, where the additional numerical noise in the calculated forces is absorbed into a Langevin-like dynamics.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

OpenACC offloading of the MFC compressible multiphase flow solver on AMD and NVIDIA GPUs

GPUs are the heart of the latest generations of supercomputers. We efficiently accelerate a compressible multiphase flow solver via OpenACC on NVIDIA and AMD Instinct GPUs. Optimization is accomplished by specifying the directive clauses gang vector and collapse. Further speedups of six and ten times are achieved by packing user-defined types into coalesced multidimensional arrays and manual inlining via metaprogramming. Additional optimizations yield seven-times speedup of array packing and thirty-times speedup of select kernels on Frontier. Weak scaling efficiencies of 97% and 95% are observed when scaling to 50% of Summit and 87% of Frontier. Strong scaling efficiencies of 84% and 81% are observed when increasing the device count by a factor of 8 and 16 on V100 and MI250X hardware. The strong scaling efficiency of AMD’s MI250X increases to 92% when increasing the device count by a factor of 16 when GPU-aware MPI is used for communication.

Wilfong, Benjamin↗

Flexible silicon photonic architecture for accelerating distributed deep learning

The increasing size and complexity of deep learning (DL) models have led to the wide adoption of distributed training methods in datacenters (DCs) and high-performance computing (HPC) systems. However, communication among distributed computing units (CUs) has emerged as a major bottleneck in the training process. In this study, we propose Flex-SiPAC, a flexible silicon photonic accelerated compute cluster designed to accelerate multi-tenant distributed DL training workloads. Flex-SiPAC takes a co-design approach that combines a silicon photonic hardware platform with a tailored collective algorithm, optimized to leverage the unique physical properties of the architecture. The hardware platform integrates a novel wavelength-reconfigurable transceiver design and a micro-resonator-based wavelength-reconfigurable switch, enabling the system to achieve flexible bandwidth steering in the wavelength domain. The collective algorithm is designed to support reconfigurable topologies, enabling efficient all-reduce communications that are commonly used in DL training. The feasibility of the Flex-SiPAC architecture is demonstrated through two testbed experiments. First, an optical testbed experiment demonstrates the flexible routing of wavelengths by shuffling an array of input wavelengths using a custom-designed spatial-wavelength selective switch. Second, a four-GPU testbed running two DL workloads shows a 23% improvement in job completion time compared to a similarly sized leaf-spine topology. We further evaluate Flex-SiPAC using large-scale simulations, which show that Flex-SiPAC is able to reduce the communication time by 26% to 29% compared to state-of-the-art compute clusters under representative collective operations.

Wu, Zhenguo (ORCID:0000000322847985)↗

A feasibility study on porting the community land model onto accelerators using OpenACC

As environmental models (such as Accelerated Climate Model for Energy (ACME), Parallel Reactive Flow and Transport Model (PFLOTRAN), Arctic Terrestrial Simulator (ATS), etc.) became more and more complicated, we are facing enormous challenges regarding to porting those applications onto hybrid computing architecture. OpenACC appears as a very promising technology, therefore, we have conducted a feasibility analysis on porting the Community Land Model (CLM), a terrestrial ecosystem model within the Community Earth System Models (CESM)). Specifically, we used automatic function testing platform to extract a small computing kernel out of CLM, then we apply this kernel into the actually CLM dataflow procedure, and investigate the strategy of data parallelization and the benefit of data movement provided by current implementation of OpenACC. Even it is a non-intensive kernel, on a single 16-core computing node, the performance (based on the actual computation time using one GPU) of OpenACC implementation is 2.3 time faster than that of OpenMP implementation using single OpenMP thread, but it is 2.8 times slower than the performance of OpenMP implementation using 16 threads. On multiple nodes, MPI_OpenACC implementation demonstrated very good scalability on up to 128 GPUs on 128 computing nodes. This study also provides useful information for us to look into the potential benefits of “deep copy” capability and “routine” feature of OpenACC standards. In conclusion, we believe that our experience on the environmental model, CLM, can be beneficial to many other scientific research programs who are interested to porting their large scale scientific code using OpenACC onto high-end computers, empowered by hybrid computing architecture.

54 ENVIRONMENTAL SCIENCES↗

Parallel accelerated vector similarity calculations for genomics applications

The surge in availability of genomic data holds promise for enabling determination of genetic causes of observed individual traits, with applications to problems such as discovery of the genetic roots of phenotypes, be they molecular phenotypes such as gene expression or metabolite concentrations, or complex phenotypes such as diseases. However, the growing sizes of these datasets and the quadratic, cubic or higher scaling characteristics of the relevant algorithms pose a serious computational challenge necessitating use of leadership scale computing. In this study we describe a new approach to performing vector similarity metrics calculations, suitable for parallel systems equipped with graphics processing units (GPUs) or Intel Xeon Phi processors. Our primary focus is the Proportional Similarity metric applied to Genome Wide Association Studies (GWAS) and Phenome Wide Association Studies (PheWAS). We describe the implementation of the algorithms on accelerated processors, methods used for eliminating redundant calculations due to symmetries, and techniques for efficient mapping of the calculations to many-node parallel systems. Finally, results are presented demonstrating high per-node performance and parallel scalability with rates of more than five quadrillion (5 × 10 15 ) elementwise comparisons achieved per second on the ORNL Titan system. In a companion paper we describe corresponding techniques applied to calculations of the Custom Correlation Coefficient for comparative genomics applications.

59 BASIC BIOLOGICAL SCIENCES↗

Automatic Offloading C++ Expression Templates to CUDA Enabled GPUs

In the last few years, many scientific applications have been developed for powerful graphics processing units (GPUs) and have achieved remarkable speedups. This success can be partially attributed to high performance host callable GPU library routines that are offloaded to GPUs at runtime. These library routines are based on C/C++-like programming toolkits such as CUDA from NVIDIA and have the same calling signatures as their CPU counterparts. Recently, with the sufficient support of C++ templates from CUDA, the emergence of template libraries have enabled further advancement in code reusability and rapid software development for GPUs. However, Expression Templates (ET), which have been very popular for implementing data parallel scientific software for host CPUs because of their intuitive and mathematics-like syntax, have been underutilized by GPU development libraries. The lack of ET usage is caused by the difficulty of offloading expression templates from hosts to GPUs due to the inability to pass instantiated expressions to GPU kernels as well as the absence of the exact form of the expressions for the templates at the time of coding. This paper presents a general approach that enables automatic offloading of C++ expression templates to CUDA enabled GPUs by using the C++ metaprogramming technique and Just-In-Time (JIT) compilation methodology to generate and compile CUDA kernels for corresponding expression templates followed by executing the kernels with appropriate arguments. This approach allows developers to port applications to run on GPUs with virtually no code modifications. More specifically, this paper uses a large ET based data parallel physics library called QDP++ as an example to illustrate many aspects of the approach to offload expression templates automatically and to demonstrate very good speedups for typical QDP++ applications running on GPUs against running on CPUs using this method of offloading. In addition, this approach of automatic offlo- ding expression templates could be applied to other many-core accelerators that provide C++ programming toolkits with the support of C++ template.

97 MATHEMATICS AND COMPUTING↗

Real-time data analysis for medical diagnosis using FPGA-accelerated neural networks

Real-time analysis of patient data during medical procedures can provide vital diagnostic feedback that significantly improves chances of success. With sensors becoming increasingly fast, frameworks such as Deep Neural Networks are required to perform calculations within the strict timing constraints for real-time operation. However, traditional computing platforms responsible for running these algorithms incur a large overhead due to communication protocols, memory accesses, and static (often generic) architectures. In this work, we implement a low-latency Multi-Layer Perceptron (MLP) processor using Field Programmable Gate Arrays (FPGAs). Unlike CPUs and Graphics Processing Units (GPUs), our FPGA-based design can directly interface sensors, storage devices, display devices and even actuators, thus reducing the delays of data movement between ports and compute pipelines. Moreover, the compute pipelines themselves are tailored specifically to the application, improving resource utilization and reducing idle cycles. We demonstrate the effectiveness of our approach using mass-spectrometry data sets for real-time cancer detection. We demonstrate that correct parameter sizing, based on the application, can reduce latency by 20% on average. Furthermore, we show that in an application with tightly coupled data-path and latency constraints, having a large amount of computing resources can actually reduce performance. Using mass-spectrometry benchmarks, we show that our proposed FPGA design outperforms both CPU and GPU implementations, with an average speedup of 144x and 21x, respectively. In our work, we demonstrate the importance of application-specific optimizations in order to minimize latency and maximize resource utilization for MLP inference. By directly interfacing and processing sensor data with ultra-low latency, FPGAs can perform real-time analysis during procedures and provide diagnostic feedback that can be critical to achieving higher percentages of successful patient outcomes.

60 APPLIED LIFE SCIENCES↗

AI-driven multiscale simulations illuminate mechanisms of SARS-CoV-2 spike dynamics

We develop a generalizable AI-driven workflow that leverages heterogeneous HPC resources to explore the time-dependent dynamics of molecular systems. We use this workflow to investigate the mechanisms of infectivity of the SARS-CoV-2 spike protein, the main viral infection machinery. Our workflow enables more efficient investigation of spike dynamics in a variety of complex environments, including within a complete SARS-CoV-2 viral envelope simulation, which contains 305 million atoms and shows strong scaling on ORNL Summit using NAMD. We present several novel scientific discoveries, including the elucidation of the spike’s full glycan shield, the role of spike glycans in modulating the infectivity of the virus, and the characterization of the flexible interactions between the spike and the human ACE2 receptor. We also demonstrate how AI can accelerate conformational sampling across different systems and pave the way for the future application of such methods to additional studies in SARS-CoV-2 and other molecular systems.

59 BASIC BIOLOGICAL SCIENCES↗

Scaling Deep Learning workloads: NVIDIA DGX-1/Pascal and Intel Knights Landing

Deep Learning (DL) algorithms have become ubiquitous in data analytics. As a result, major computing vendors—including NVIDIA, Intel, AMD, and IBM—have architectural road maps influenced by DL workloads. Furthermore, several vendors have recently advertised new computing products as accelerating large DL workloads. Unfortunately, it is difficult for data scientists to quantify the potential of these different products. Here, this article provides a performance and power analysis of important DL workloads on two major parallel architectures: NVIDIA DGX-1 (eight Pascal P100 GPUs interconnected with NVLink) and Intel Knights Landing (KNL) CPUs interconnected with Intel Omni-Path or Cray Aries. Our evaluation consists of a cross section of convolutional neural net workloads: CifarNet, AlexNet, GoogLeNet, and ResNet50 topologies using the Cifar10 and ImageNet datasets. The workloads are vendor-optimized for each architecture. We use sequentially equivalent implementations to maintain iso-accuracy between parallel and sequential DL models. Our analysis indicates that although GPUs provide the highest overall performance, the gap can close for some convolutional networks; and the KNL can be competitive in performance/watt. We find that NVLink facilitates scaling efficiency on GPUs. However, its importance is heavily dependent on neural network architecture. Furthermore, for weak-scaling—sometimes encouraged by restricted GPU memory—NVLink is less important.

97 MATHEMATICS AND COMPUTING↗

Development and Benchmarking of a Hybrid PIC Code For Dense Plasmas and Fast Ignition (Final Report)

Radiation processes play an important role in the study of both fast ignition and other inertial confinement schemes, such as plasma jet driven magneto-inertial fusion, both in their effect on energy balance, and in generating diagnostic signals. In the latter case, warm and hot dense matter may be produced by the convergence of a plasma shell formed by the merging of an assembly of high Mach number plasma jets. This innovative approach has the potential advantage of creating matter of high energy densities in voluminous amount compared with high power lasers or particle beams. An important application of this technology is as a plasma liner for the flux compression of magnetized plasma to create ultra-high magnetic fields and burning plasmas. HyperV Technologies Corp. has been developing plasma jet accelerator technology in both coaxial and linear railgun geometries to produce plasma jets of sufficient mass, density, and velocity to create such imploding plasma liners. An enabling tool for the development of this technology is the ability to model the plasma dynamics, not only in the accelerators themselves, but also in the resulting magnetized target plasma and within the merging/interacting plasma jets during transport to the target. Welch pioneered numerical modeling of such plasmas (including for fast ignition) using the LSP simulation code. Lsp is an electromagnetic, parallelized, plasma simulation code under development since 1995. It has a number of innovative features making it uniquely suitable for modeling high energy density plasmas including a hybrid fluid model for electrons that allows electrons in dense plasmas to be modeled with a kinetic or fluid treatment as appropriate. In addition to in-house use at Voss Scientific, several groups carrying out research in Fast Ignition (LLNL, SNL, UCSD, AWE (UK), and Imperial College (UK)) also use LSP. A collaborative team consisting of HyperV Technologies Corp., Voss Scientific LLC, FAR-TECH, Inc., Prism Computational Sciences, Inc. and Advanced Energy Systems Inc. joined efforts to develop new physics and numerical models for LSP in several key areas to enhance the ability of LSP to model high energy density plasmas (HEDP). This final report details those efforts. Areas addressed in this research effort include: adding radiation transport to LSP, first in 2D and then fully 3D, extending the EMHD model to 3D, implementing more advanced radiation and electrode plasma boundary conditions, and installing more efficient implicit numerical algorithms to speed complex 2-D and 3-D computations. The new capabilities allow modeling of the dominant processes in high energy density plasmas, and further assist the development and optimization of plasma jet accelerators, with particular attention to MHD instabilities and plasma/wall interaction (based on physical models for ion drag friction and ablation/erosion of the electrodes). In the first funding cycle we implemented a solver for the radiation diffusion equation. To solve this equation in 2-D, we used finite-differencing and applied the parallelized sparse-matrix solvers in the PETSc library (Argonne National Laboratory) to the resulting system of equations. A database of the necessary coefficients for materials of interest was assembled using the PROPACEOS and ATBASE codes from Prism. The model was benchmarked against Prism's 1-D radiation hydrodynamics code HELIOS, and against experimental data obtained from HyperV's separately funded plasma jet accelerator development program. Work in the second funding cycle focused on extending the radiation diffusion model to full 3-D, continued development of the EMHD model, optimizing the direct-implicit model to speed up calculations, add in multiply ionized atoms, and improved the way boundary conditions are handled in LSP. These new LSP capabilities were then used, along with analytic calculations and Mach2 runs, to investigate plasma jet merging, plasma detachment and transport, restrike and advanced jet accelerator design. In addition, a strong linkage to diagnostic measurements was made by modeling plasma jet experiments on PLX to support benchmarking of the code. A large number of upgrades and improvements advancing hybrid PIC algorithms were implemented in LSP during the second funding cycle. These include development of fully 3D radiation transport algorithms, new boundary conditions for plasma-electrode interactions, and a charge conserving equation of state that permits multiply ionized high-Z ions. The final funding cycle focused on 1) mitigating the effects of a slow-growing grid instability which is most pronounced in plasma jet frame expansion problems using the two-fluid Eulerian remap algorithm, 2) extension of the Eulerian Smoothing Algorithm to allow EOS/Radiation modeling, 3) simulations of collisionless shocks formed by jet merging, 4) simulations of merging jets using high-Z gases, 5) generation of PROPACEOS EOS/Opacity databases, 6) simulations of plasma jet transport experiments, 7) simulations of plasma jet penetration through transverse magnetic fields, and 8) GPU PIC code development. The tools developed during this project are applicable not only to the study of plasma jets, but also to a wide variety of HEDP plasmas of interest to DOE, including plasmas created in short-pulse laser experiments performed to study fast ignition concepts for inertial confinement fusion.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Analysis of GPU Data Access Patterns on Complex Geometries for the D3Q19 Lattice Boltzmann Algorithm

GPU performance of the lattice Boltzmann method (LBM) depends heavily on memory access patterns. When implemented with GPUs on complex domains, typically, geometric data is accessed indirectly and lattice data is accessed lexicographically. Although there are a variety of other options, no study has examined the relative efficacy between them. Here, we examine a suite of memory access schemes via empirical testing and performance modeling. We find strong evidence that semi-direct is often better suited than the more common indirect addressing, providing increased computational speed and reducing memory consumption. For the layout, we find that the Collected Structure of Arrays (CSoA) and bundling layouts outperform the common Structure of Array layout; on V100 and P100 devices, CSoA consistently outperforms bundling, however the relationship is more complicated on K40 devices. When compared to state-of-the-art practices, our recommendations lead to speedups of 10–40 percent and reduce memory consumption up to 17 percent. Using performance modeling and computational experimentation, we determine the mechanisms behind the accelerations. We demonstrate that our results hold across multiple GPUs on two leadership class systems, and present the first near-optimal strong results for LBM with arterial geometries run on GPUs.

42 ENGINEERING↗

Accelerating Subsurface Transport Simulation on Heterogeneous Clusters

Reactive transport numerical models simulate chemical and microbiological reactions that occur along a flowpath. These models have to compute reactions for a large number of locations. They solve the set of ordinary differential equations (ODEs) that describes the reaction for each location through the Newton-Raphson technique. This technique involves computing a Jacobian matrix and a residual vector for each set of equation, and then solving iteratively the linearized system by performing Gaussian Elimination and LU decomposition until convergence. STOMP, a well known subsurface flow simulation tool, employs matrices with sizes in the order of 100x100 elements and, for numerical accuracy, LU factorization with full pivoting instead of the faster partial pivoting. Modern high performance computing systems are heterogeneous machines whose nodes integrate both CPUs and GPUs, exposing unprecedented amounts of parallelism. To exploit all their computational power, applications must use both the types of processing elements. For the case of subsurface flow simulation, this mainly requires implementing efficient batched LU-based solvers and identifying efficient solutions for enabling load balancing among the different processors of the system. In this paper we discuss two approaches that allows scaling STOMP's performance on heterogeneous clusters. We initially identify the challenges in implementing batched LU-based solvers for small matrices on GPUs, and propose an implementation that fulfills STOMP's requirements. We compare this implementation to other existing solutions. Then, we combine the batched GPU solver with an OpenMP-based CPU solver, and present an adaptive load balancer that dynamically distributes the linear systems to solve between the two components inside a node. We show how these approaches, integrated into the full application, provide speed ups from 6 to 7 times on large problems, executed on up to 16 nodes of a cluster with two AMD Opteron 6272 and a Tesla M2090 per node.

GPU↗

A Case Study with the HACCmk Kernel in SYCL

The SYCL standard specifies a cross-platform abstraction layer that enables programming of heterogeneous computing systems using standard C++ [1]. In the Open Computing Language (OpenCL) programming model, host and device codes are written in different languages [2 ]. To improve development productivity and performance portability, the SYCL programming model combines host and device codes for an application in a type-safe way. In this report, we are interested in applying the SYCL programming model to a computationally intensive routine derived from the Hardware Accelerated Cosmology Code (HACC) framework for a study on performance portability on a heterogeneous computing device. We use Intel® OneAPI toolkit [3] to build the OpenCL and SYCL programs, and evaluate the performance of both implementations on Intel® integrated GPUs. We find that the SYCL implementation can achieve the same performance as the OpenCL implementation after we specify a target backend for the compiler to build the SYCL program in offline compilation. Without the specification of a target backend, the runtime will have to compile intermediate device codes for a target platform. Such runtime overhead may become significant compared to the execution of a kernel on a target device. As the kernel routine is compute-bound, we evaluate the impact of the number of compute units upon the kernel’s raw performance using two GPUs. When the hardware resource is not underutilized, we can obtain almost linear performance speedup from 48 compute units to 72 compute units in offline compilation, which indicates that the number of compute units on a GPU are important to improving the raw performance of a compute-bound kernel. The experimental results show that SYCL is a promising programming model for heterogeneous computing. The remainder of the report is organized as follows: In Section II, we describe the SYCL programming model, the steps to map an OpenCL program to a SYCL program, and the architecture of an integrated GPU. Section III introduces the kernel routine, and presents the OpenCL and SYCL implementations of the kernel. In Section IV, we evaluate the performance of the implementations on the GPUs. Section V concludes the report.

97 MATHEMATICS AND COMPUTING↗

Active Storage with Analytics Capabilities and I/O Runtime System for Petascale Systems

Computational scientists must understand results from experimental, observational and computational simulation generated data to gain insights and perform knowledge discovery. As systems approach the petascale range, problems that were unimaginable a few years ago are within reach. With the increasing volume and complexity of data produced by ultra-scale simulations and high-throughput experiments, understanding the science is largely hampered by the lack of comprehensive I/O, storage, acceleration of data manipulation, analysis, and mining tools. Scientists require techniques, tools and infrastructure to facilitate better understanding of their data, in particular the ability to effectively perform complex data analysis, statistical analysis and knowledge discovery. The goal of this work is to enable more effective analysis of scientific datasets through the integration of enhancements in the I/O stack, from active storage support at the file system layer to MPI-IO and high-level I/O library layers. We propose to provide software components to accelerate data analytics, mining, I/O, and knowledge discovery for large-scale scientific applications, thereby increasing productivity of both scientists and the systems. Our approaches include 1) design the interfaces in high-level I/O libraries, such as parallel netCDF, for applications to activate data mining operations at the lower I/O layers; 2) Enhance MPI-IO runtime systems to incorporate the functionality developed as a part of the runtime system design; 3) Develop parallel data mining programs as part of runtime library for server-side file system in PVFS file system; and 4) Prototype an active storage cluster, which will utilize multicore CPUs, GPUs, and FPGAs to carry out the data mining workload.

97 MATHEMATICS AND COMPUTING↗

FPDeep: Scalable Acceleration of CNN Training on Deeply-Pipelined FPGA Clusters

In this paper, we propose a framework called FPDeep, which uses a hybrid of model and layer paral- lelism to configure distributed reconfigurable clusters to train DNNs. This approach has numerous benefits. First, the design does not suffer from batch size growth. Second, novel workload and weight partitioning leads to balanced loads of both among nodes. And third, the entire system is fine-grained pipeline. This leads to high parallelism and utilization and also minimizes the time features need to be cached while waiting for back-propagation.

Wang, Tianqi↗