Search NASA⌕ Search

SEARCH · Search NASA

Results for “Memory Bandwidth”

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 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
NASA NTRS2005

An Application-Based Performance Characterization of the Columbia Supercluster

Columbia is a 10,240-processor supercluster consisting of 20 Altix nodes with 512 processors each, and currently ranked as the second-fastest computer in the world. In this paper, we present the performance characteristics of Columbia obtained on up to four computing nodes interconnected via the InfiniBand and/or NUMAlink4 communication fabrics. We evaluate floating-point performance, memory bandwidth, message passing communication speeds, and compilers using a subset of the HPC Challenge benchmarks, and some of the NAS Parallel Benchmarks including the multi-zone versions. We present detailed performance results for three scientific applications of interest to NASA, one from molecular dynamics, and two from computational fluid dynamics. Our results show that both the NUMAlink4 and the InfiniBand hold promise for application scaling to a large number of processors.

Biswas, Rupak↗
NASA NTRS2015

High-Speed On-Board Data Processing for Science Instruments: HOPS

The project called High-Speed On-Board Data Processing for Science Instruments (HOPS) has been funded by NASA Earth Science Technology Office (ESTO) Advanced Information Systems Technology (AIST) program during April, 2012 â€" April, 2015. HOPS is an enabler for science missions with extremely high data processing rates. In this three-year effort of HOPS, Active Sensing of CO2 Emissions over Nights, Days, and Seasons (ASCENDS) and 3-D Winds were of interest in particular. As for ASCENDS, HOPS replaces time domain data processing with frequency domain processing while making the real-time on-board data processing possible. As for 3-D Winds, HOPS offers real-time high-resolution wind profiling with 4,096-point fast Fourier transform (FFT). HOPS is adaptable with quick turn-around time. Since HOPS offers reusable user-friendly computational elements, its FPGA IP Core can be modified for a shorter development period if the algorithm changes. The FPGA and memory bandwidth of HOPS is 20 GB/sec while the typical maximum processor-to-SDRAM bandwidth of the commercial radiation tolerant high-end processors is about 130-150 MB/sec. The inter-board communication bandwidth of HOPS is 4 GB/sec while the effective processor-to-cPCI bandwidth of commercial radiation tolerant high-end boards is about 50-75 MB/sec. Also, HOPS offers VHDL cores for the easy and efficient implementation of ASCENDS and 3-D Winds, and other similar algorithms. A general overview of the 3-year development of HOPS is the goal of this presentation.

Beyon, Jeffrey↗
NASA NTRS2019

Application of Machine Learning Techniques to Delay Tolerant Network Routing

This dissertation discusses several machine learning techniques to improve routing in delay tolerant networks (DTNs). These are networks in which there may be long one-way trip times, asymmetric links, high error rates, and deterministic as well as non-deterministic loss of contact between network nodes, such as interplanetary satellite networks, mobile ad hoc networks and wireless sensor networks. This work uses historical network statistics to train a multi-label classifier to predict reliable paths through the network. In addition, a clustering technique is used to predict future mobile node locations. Both of these techniques are used to reduce the consumption of resources such as network bandwidth, memory and data storage that is required by replication routing methods often used in opportunistic DTN environments. Thesis contributions include: an emulation tool chain developed to create a DTN test bed for machine learning, the network and software architecture for a machine learning based routing method, the development and implementation of classification and clustering techniques and performance evaluation in terms of machine learning and routing metrics.

Dudukovich, Rachel M.↗
NASA NTRSDate not supplied

NASA and Blue Origin Collaborative Assessment of Precision Landing Algorithms and Computing

NASA’s Safe and Precise Landing Integrated Capabilities Evolution (SPLICE) project is developing sensor, algorithm, and compute technologies for precision landing and hazard avoidance. These technologies are being tested as an integrated Precision Landing and Hazard Avoidance (PL&HA) system on Blue Origin’s New Shephard suborbital vehicle. A key goal for the computing element of this technology development is to characterize the performance of the SPLICE software workloads on the project’s Descent and Landing Computer (DLC). The DLC is a multi-core processor designed as a surrogate for NASA’s High-Performance Space Computer (HPSC). Measurements of the SPLICE workload performance on the DLC provides NASA insight on how PL&HA capabilities will perform on the HPSC, and guidance on how the SPLICE algorithms can be implemented to best utilize the DLC platform. This insight can also be used to derive requirements to guide trade studies on candidate computing architectures, for use on platforms like Blue Moon. NASA and Blue Origin are collaborating under an agreement to pursue this mutual benefit. Performance metrics collected are based on measurement of common compute resources such as percentage used of memory bandwidth, I/O utilization, interrupt latency, and kernel vs. user space code residency. Where possible existing performance counters and metrics that are part of the operating system kernel are used. As the design has a significant FPGA component, performance counters are identified and instantiated in the fabric to measure DMA performance and interface metrics. Collection of metrics is performed on the DLC with a representative workload that simulates a full landing cycle of the Blue Origin New Shepard vehicle. Consideration is given to the other compute implementations and whether they can run SPLICE algorithms at the same rate and with the same latency as the DLC. One option being considered is the use of a RISC-V soft core instantiated in a radiation resilient FPGA fabric such as the Xilinx KU60. Select algorithms from the SPLICE code will be run for comparison with the DLC. This paper describes how the DLC is instrumented to collect performance measurements of the SPLICE workloads, preliminary results from these measurements, and their implications on SPLICE algorithm implementation. The results of experimentation to derive candidate requirements for architecture trades on a PL&HA computing system are also presented.

computer performance↗
NASA NTRSDate not supplied

Improvements to a Batch Pentadiagonal Solver on NVIDIA GPUs

This poster presents the recent work in OVERFLOW to port the batched pentadiagonal solver to NVIDIA GPUs. There are five pentadiagonal systems for each pencil in the grid but three of these systems share the same LHS. Our first simple approach for porting the pentadiagonal solver to the GPUs was to take advantage of the shared LHS by assigning three threads to the three LHS of each pencil. We demonstrated that this custom solver was 92% faster than the NVIDIA batched pentadiagonal library implementation on a V100 GPU due to the lower memory bandwidth requirements. The second approach treated each pentadiagonal system as a 2x2 block tridiagonal system and used a variant of the parallel cyclic reduction algorithm to solve the problem. One benefit of this approach is that it does not require interleaving the data between each system. We demonstrated that this algorithm is 2.18x faster than the NVIDIA library implementation for the same amount of work. If we take advantage of our shared LHS, this approach is 2.58x faster than the library implementation on a V100 GPU.

GPU Programming↗
DOE OSTI2016

MILC staggered conjugate gradient performance on Intel KNL

We review our work done to optimize the staggered conjugate gradient (CG) algorithm in the MILC code for use with the Intel Knights Landing (KNL) architecture. KNL is the second gener- ation Intel Xeon Phi processor. It is capable of massive thread parallelism, data parallelism, and high on-board memory bandwidth and is being adopted in supercomputing centers for scientific research. The CG solver consumes the majority of time in production running, so we have spent most of our effort on it. We compare performance of an MPI+OpenMP baseline version of the MILC code with a version incorporating the QPhiX staggered CG solver, for both one-node and multi-node runs.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Revisiting Huffman Coding: Toward Extreme Performance on Modern GPU Architectures

Today's high-performance computing (HPC) applications are producing vast volumes of data, which are challenging to store and transfer efficiently during the execution, such that data compression is becoming a critical technique to mitigate the storage burden and data movement cost. Huffman coding is arguably the most efficient Entropy coding algorithm in information theory, such that it could be found as a fundamental step in many modern compression algorithms such as DEFLATE. On the other hand, today's HPC applications are more and more relying on the accelerators such as GPU on supercomputers, while Huffman encoding suffers from low throughput on GPUs, resulting in a significant bottleneck in the entire data processing. In this paper, we propose and implement an efficient Huffman encoding approach based on modern GPU architectures, which addresses two key challenges: (1) how to parallelize the entire Huffman encoding algorithm, including codebook construction, and (2) how to fully utilize the high memory-bandwidth feature of modern GPU architectures. The detailed contribution is fourfold. (1) We develop an efficient parallel codebook construction on GPUs that scales effectively with the number of input symbols. (2) We propose a novel reduction based encoding scheme that can efficiently merge the codewords on GPUs. (3) We optimize the overall GPU performance by leveraging the state-of-the-art CUDA APIs such as Cooperative Groups. (4) We evaluate our Huffman encoder thoroughly using six real-world application datasets on two advanced GPUs and compare with our implemented multithreaded Huffman encoder. Experiments show that our solution can improve the encoding throughput by up to 5.0× and 6.8× on NVIDIA RTX 5000 and V100, respectively, over the state-of-the-art GPU Huffman encoder, and by up to 3.3× over the multithread encoder on two 28-core Xeon Platinum 8280 CPUs.

Tian, Jiannan↗
DOE OSTI2020

OpenACC unified programming environment for GPU and FPGA multi-hybrid acceleration

Attached accelerators have been frequently used in recent High Per- formance Computing (HPC) systems because of their high performance/power ratio. In particular, the Graphics Processing Unit (GPU) is the most popu- lar accelerator owing to its high peak FLOPS performance and high memory bandwidth supported by HBM2, etc. However, the performance of GPU depends highly on a large degree of SIMD parallelism and has difficulty sustaining a high performance on programs with frequent branch operations or a partially low degree of parallelism.By contrast, a Field Programmable Gate Array (FPGA) has received attention as a different type of accelerator than GPU as a fully reconfigurable processor fitting the target applications. The high performance of FPGA is mainly provided by a pipelined operation and optimized circuit suitable for any operation even with frequent conditional branches. We have been focusing on the flexibility of FPGA to compensate for the weakness of GPU. We believe that the coupling of GPU with FPGA can result in one of the most powerful accelerating platforms available.However, the program coding of GPU and FPGA coupling can be quite difficult for application users. Traditionally, CUDA by NVIDIA has been the most popular programming language with the largest share of GPUs used in HPC, whereas a hardware description language such as Verilog HDL has been used in FPGA programming. OpenCL coding has recently become available even on high-end FPGAs. Moreover, several recent studies have also enabled the OpenACC coding for use in FPGA. In this study, we provide a unified programming system based on OpenACC for a platform equipped with both GPU and FPGA aiming at the next-generation accelerated supercomputer framework. Our programming environment is called Multi-Hybrid OpenACC Translator (MHOAT), and in this paper, we describe the basic concept and prototype system of MHOAT based on an evaluation on the amount of coding required and the performance of a hybrid multi-device accelerated system.

Tsunashima, Ryuta↗
DOE OSTI2016

GPU Acceleration of History-Based Multigroup Monte Carlo

Accurate solutions of the radiation transport equation play an important role in the analysis of many nuclear systems. Because there is no need to discretize the problem, Monte Carlo methods offer the most accurate possible solutions to the transport equation. However, significant computational effort is required to reduce the stochastic noise inherent to Monte Carlo methods. Current trends in high performance computing are moving towards vectorized, single instruction multiple data (SIMD), architectures such as Intel Xeon Phis and graphics processing units (GPUs) due to their favorable ratio of performance to power consumption. The challenge of adapting Monte Carlo transport to vectorized computing architectures was first addressed by Brown with an algorithm known as event-based Monte Carlo. In this approach, rather than simulating individual particle histories from birth to death, groups of particles are processed one event at a time. This allows each event batch to perform the same operations in lockstep, enabling use of the vector hardware. Although modern vectorized hardware is more versatile than the early vector machines, reducing so-called thread divergence is nonetheless still an important consideration for algorithm design. Many of the algorithms proposed so far for Monte Carlo transport on advanced architectures have focused on variants of event-based transport. Some evidence, however, suggests that low-level thread divergence, resulting from short-lived branching statements, may not be detrimental to overall performance on GPUs as long as the effect on achievable memory bandwidth is considered. This paper characterizes the performance of a traditional history-based Monte Carlo algorithm on GPUs. We show that with only minor modifications to traditional algorithms, significant performance gains relative to CPUs are possible. (authors)

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2017

CUDA Computation of the Feynman Distribution

In 2006, NVIDIA Corporation introduced the Computer Unified Device Architecture (CUDA) for C, C++, and Fortran programming languages. The CUDA parallel computing platform uses the Graphics Processing Unit (GPU) of NVIDIA graphics card, and it is designed for parallel computations. In these computations, the same algorithms are executed on the data in parallel. The GPU contains many Streaming Multiprocessors (SM) that can independently execute blocks of threads. A thread is a sequence of serial programmed instructions. The SM executes threads in groups of 32 and a group of 32 threads is called warp. The threads within a block can be synchronized and have access to a shared memory that is visible only to them. Conversely, blocks of threads cannot be synchronized and have access only to the global memory of the graphics card. This memory is called device memory and is independent of the host memory controlled by the Central Processing Unit (CPU). A single thread has access to a local memory (register) that is not visible to any other thread. The part of the program that is executed on the graphics card is called kernel. Each thread of the kernel has a unique identification index that is accessible by the in-built variable threadIdx; the latter is a 3-component vector. A threads block can contain up to 1024 threads. Blocks of threads have a unique identification index that is accessible by the in-built variable blockIdx; the latter is a 3-component vector. The in-built variable blockDim contains the number of threads per block; this variable is a 3-component vector. Threads and blocks represent different levels of parallelism; a kernel is executed as a grid of blocks of threads. Figure 1 illustrates a grid of thread blocks for a GPU with 2 SM. A SM can execute multiple blocks of threads in parallel; a block of threads is executed on a single SM. A GPU can execute multiple kernels concurrently. This work has used the Quadro M5000 graphics card with the following technical specifications: - 2048 cores with 861 MHz frequency, - 16 streaming multiprocessors, - 5.2 trillion transistors, - 8.2 Gb device memory, - 64 kb register memory (per thread), - 48 kb shared memory (per block) - 211.6 GB/s memory bandwidth, and - 3.5 GFLOPS floating point performance.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Sum Reduction with OpenMP Offload on NVIDIA Grace-Hopper System

We evaluate the performance of the baseline and optimized reductions in OpenMP on an NVIDIA Grace-Hopper system. We explore the impacts of the number of teams, the number of elements to sum per loop iteration, and simultaneous execution on the central-processing unit (CPU) and the GPU in the unified memory (UM) mode upon the reduction performance. The experimental results show that the optimized reductions are 6.120X to 20.906X faster than the baselines on the GPU, and their efficiency ranges from 89% to 95% of the theoretical GPU memory bandwidth. Depending on where an input array is allocated in the program when co-running the reduction on the CPU and GPU in the UM mode, the average speedup over the GPU-only execution is approximately 2.484 or 1.067, and the speedup of the optimized reductions over the baseline reductions ranges from 0.996 to 10.654 or from 0.998 to 6.729.

Jin, Zheming↗
NASA NTRS2008

High-Density, High-Bandwidth, Multilevel Holographic Memory

A proposed holographic memory system would be capable of storing data at unprecedentedly high density, and its data transfer performance in both reading and writing would be characterized by exceptionally high bandwidth. The capabilities of the proposed system would greatly exceed even those of a state-of-the art memory system, based on binary holograms (in which each pixel value represents 0 or 1), that can hold .1 terabyte of data and can support a reading or writing rate as high as 1 Gb/s. The storage capacity of the state-of-theart system cannot be increased without also increasing the volume and mass of the system. However, in principle, the storage capacity could be increased greatly, without significantly increasing the volume and mass, if multilevel holograms were used instead of binary holograms. For example, a 3-bit (8-level) hologram could store 8 terabytes, or an 8-bit (256-level) hologram could store 256 terabytes, in a system having little or no more size and mass than does the state-of-the-art 1-terabyte binary holographic memory. The proposed system would utilize multilevel holograms. The system would include lasers, imaging lenses and other beam-forming optics, a block photorefractive crystal wherein the holograms would be formed, and two multilevel spatial light modulators in the form of commercially available deformable-mirror-device spatial light modulators (DMDSLMs) made for use in high speed input conversion of data up to 12 bits. For readout, the system would also include two arrays of complementary metal oxide/semiconductor (CMOS) photodetectors matching the spatial light modulators. The system would further include a reference-beam sterring device (equivalent of a scanning mirror), containing no sliding parts, that could be either a liquid-crystal phased-array device or a microscopic mirror actuated by a high-speed microelectromechanical system. Time-multiplexing and the multilevel nature of the DMDSLM would be exploited to enable writing and reading of multilevel holograms. The DMDSLM would also enable transfer of data at a rate of 7.6 Gb/s or perhaps somewhat higher.

Chao, Tien-Hsin↗
NASA NTRS1980

Expanded interleaved solid-state memory for a wide bandwidth transient waveform recorder

An interleaved, solid state expanded memory for a 100 MHz bandwidth waveform recorder is described. The memory development resulted in a significant increase in the storage capacity of a commercially available recorder. The motivation for the memory expansion of the waveform recorder, which is used to support in-flight measurement of the electromagnetic characteristics of lightning discharges, was the need for a significantly longer data window than that provided by the commercially available unit. The expanded recorder provides a data window that is 128 times longer than the commercial unit, while maintaining the same time resolution, by increasing the storage capacity from 1024 to 131 072 data samples. The expanded unit operates at sample periods as small as 10 ns. Sampling once every 10 ns, the commercial unit records for about 10 microseconds before the memory is filled, whereas, the expanded unit records for about 1300 microseconds. A photo of the expanded waveform recorder is shown.

Thomas, R. M., Jr.↗
NASA NTRS1978

A wide bandwidth CCD buffer memory system

A prototype system was implemented to demonstrate that CCD's can be applied advantageously to the problem of low power digital storage and particularly to the problem of interfacing widely varying data rates. CCD shift register memories (8K bit) were used to construct a feasibility model 128 K-bit buffer memory system. Serial data that can have rates between 150 kHz and 4.0 MHz can be stored in 4K-bit, randomly-accessible memory blocks. Peak power dissipation during a data transfer is less than 7 W, while idle power is approximately 5.4 W. The system features automatic data input synchronization with the recirculating CCD memory block start address. System expansion to accommodate parallel inputs or a greater number of memory blocks can be performed in a modular fashion. Since the control logic does not increase proportionally to increase in memory capacity, the power requirements per bit of storage can be reduced significantly in a larger system.

Siemens, K.↗
NASA NTRS2006

Autonomous Detection of Dust Devils and Clouds on Mars

Acquisition of science in space applications is shifting from teleoperated gathering to an automated on-board analysis with improvements in the use of on-board memory, CPU, bandwidth and data quality. In this paper, we describe algorithms to autonomously detect dust devils and clouds from a rover and summarize the results. These algorithms meet high hit-to-miss ratios and satisfy strict requirements of CPU, memory usage and bandwidth. The detectors have been scheduled for upload to the Mars Exploration Rovers (MER) in 2006. These are the first autonomous science processes in the rovers.

onboard science↗
DOE OSTI2024

Tailor : Altering Skip Connections for Resource-Efficient Inference

Deep neural networks use skip connections to improve training convergence. However, these skip connections are costly in hardware, requiring extra buffers and increasing on- and off-chip memory utilization and bandwidth requirements. In this article, we show that skip connections can be optimized for hardware when tackled with a hardware-software codesign approach. We argue that while a network’s skip connections are needed for the network to learn, they can later be removed or shortened to provide a more hardware-efficient implementation with minimal to no accuracy loss. We introduceTailor, a codesign tool whose hardware-aware training algorithm gradually removes or shortens a fully trained network’s skip connections to lower the hardware cost.Tailorimproves resource utilization by up to 34% for block random access memories (BRAMs), 13% for flip-flops (FFs), and 16% for look-up tables (LUTs) for on-chip, dataflow-style architectures.Tailorincreases performance by 30% and reduces memory bandwidth by 45% for a two-dimensional processing element array architecture.

Computer Science↗
NASA NTRS1986

Application of data flow concepts to a multigrid solver for the Euler equations

In this study a multigrid solver for Euler equations (FLO52R) was examined to determine its performance potential on a hypothetical computer using a data flow architecture. The proposed computer would require massive parallelism to realize its design performance. On the other hand this parallelism would be more easily realized than with a conventional vector processor such as the Cray-1S. Several changes to the proposed design substantially alleviated most of the remaining bottlenecks to parallel processing. Other changes allowed clearer definition of memory access and disk I/O. Finally, a portion of the algorithm was rewritten to improve parallel performance. With these changes, performance levels approaching that of a Cray-1S may be possible for a computer costing far less. Estimates are given for overall speed, memory, and network bandwidth, and for instruction memory requirements.

Merriam, M. L.↗
DOE OSTI2017

Preparing NERSC users for Cori, a Cray XC40 system with Intel many integrated cores

The newest NERSC supercomputer Cori is a Cray XC40 system consisting of 2,388 Intel Xeon Haswell nodes and 9,688 Intel Xeon-Phi “Knights Landing” (KNL) nodes. Compared to the Xeon-based clusters NERSC users are familiar with, optimal performance on Cori requires consideration of KNL mode settings; process, thread, and memory affinity; fine-grain parallelization; vectorization; and use of the high-bandwidth MCDRAM memory. This paper describes our efforts preparing NERSC users for KNL through the NERSC Exascale Science Application Program, Web documentation, and user training. We discuss how we configured the Cori system for usability and productivity, addressing programming concerns, batch system configurations, and default KNL cluster and memory modes. Here, system usage data, job completion analysis, programming and running jobs issues, and a few successful user stories on KNL are presented.

97 MATHEMATICS AND COMPUTING↗