Search NASA⌕ Search

SEARCH · Search NASA

Results for “memory efficiency”

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

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

At least 1,279 records · Page 71

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↗

A fast band–Krylov eigensolver for macromolecular functional motion simulation on multicore architectures and graphics processors

We introduce a new iterative Krylov subspace-based eigensolver for the simulation of macromolecular motions on desktop multithreaded platforms equipped with multicore processors and, possibly, a graphics accelerator (GPU). The method consists of two stages, with the original problem first reduced into a simpler band-structured form by means of a high-performance compute-intensive procedure. This is followed by a memory-intensive but low-cost Krylov iteration, which is off-loaded to be computed on the GPU by means of an efficient data-parallel kernel. The experimental results reveal the performance of the new eigensolver. Concretely, when applied to the simulation of macromolecules with a few thousands degrees of freedom and the number of eigenpairs to be computed is small to moderate, the new solver outperforms other methods implemented as part of high-performance numerical linear algebra packages for multithreaded architectures.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Hysteretic Responses of Skyrmion Lattices to Electric Fields in Magnetoelectric Cu 2 OSeO 3

Electric field control of topologically nontrivial magnetic textures, such as skyrmions, provides a paradigm shift for future spintronics beyond current silicon-based technology. While significant progress has been made by X-ray and neutron scattering studies, direct observation of such nanoscale spin structures and their dynamics driven by external electric fields remains a challenge to understand underlying mechanisms and harness functionalities. Here, using Lorentz transmission electron microscopy combined with in situ electric and magnetic fields at liquid helium temperatures, we report the crystallographic-orientation dependent skyrmion responses to electric fields in thin slabs of magnetoelectric Cu 2 OSeO 3 . We show electric fields not only stabilize the hexagonally packed skyrmion lattices in the entire sample in a hysteretic manner but also induce the rotation of their reciprocal vector discretely by 30°. Furthermore, the nonvolatile and energy-efficient skyrmion-lattice control by electric fields demonstrated in this work provides an important foundation for designing skyrmion-based qubits and memory devices.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Evolution of the ATLAS event data model for the HL-LHC

The upcoming high-luminosity run of the CERN Large Hadron Collider (HL-LHC) will yield an unprecedented volume of data. In order to process this data, the ATLAS collaboration is evolving its offline software to be able to use heterogeneous resources such as graphical processing units (GPUs) and field-programmable gate arrays (FPGAs). To reduce conversion overheads, the event data model (EDM) should be compatible with the requirements of these resources. While the ATLAS EDM has long allowed representing data as a structure of arrays, further evolution of the EDM can enable more efficient sharing of data between CPU and GPU resources. Some of this work will be summarized here, including extensions to allow controlling how memory for event data is allocated and the implementation of jagged vectors.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Analyzing the Impact of System Reliability Events on Applications in the Titan Supercomputer

Extreme-scale computing systems employ Reliability, Availability and Serviceability (RAS) mechanisms and infrastructure to log events from multiple system components. In this paper, we analyze RAS logs in conjunction with the application placement and scheduling database, in order to understand the impact of common RAS events on application performance. This study conducted on the records of about 2 million applications executed on Titan supercomputer provides important insights for system users, operators and computer science researchers. Specifically, we investigate the impact of RAS events on application performance and its variability by comparing cases where events are recorded with corresponding cases where no events are recorded. Such a statistical investigation is possible since we observed that system users tend to execute their applications multiple times. Our analysis reveals that most RAS events do impact application performance, although not always. We also find that different system components affect application performance differently. In particular, our investigation includes the following components: parallel file system processor, memory, graphics processing units, system and user software issues. Our work establishes the importance of providing feedback to system users for increasing operational efficiency of extreme-scale systems.

Ashraf, Rizwan↗

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. This paper 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. 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.

Gawande, Nitin A.↗

Rapid Calculation of Max-Min Fair Rates for Multi-Commodity Flows in Fat-Tree Networks

Max-min fairness is often used in the performance modeling of interconnection networks. Existing methods to compute max-min fair rates for multi-commodity flows have high complexity and are computationally infeasible for large networks. In this paper, we show that by considering topological features, this problem can be solved efficiently for the fat-tree topology that is widely used in data centers and high performance compute clusters. Several efficient new algorithms are developed for this problem, including a parallel algorithm that can take advantage of multi-core and shared-memory architectures. Using these algorithms, we demonstrate that it is possible to find the max-min fair rate allocation for multi-commodity flows in fat-tree networks that support tens of thousands of nodes. We evaluate the run-time performance of the proposed algorithms and show improvement in orders of magnitude over the previously best known method. Finally, we further demonstrate a new application of max-min fair rate allocation that is only computationally feasible using our new algorithms.

97 MATHEMATICS AND COMPUTING↗

Deep quantum circuit simulations of low-energy nuclear states

Numerical simulation is an important method for verifying the quantum circuits used to simulate low-energy nuclear states. However, real-world applications of quantum computing for nuclear theory often generate deep quantum circuits that place demanding memory and processing requirements on conventional simulation methods. Here, we present advances in high-performance numerical simulations of deep quantum circuits to efficiently verify the accuracy of low-energy nuclear physics applications. Our approach employs novel methods for accelerating the numerical simulation including management of simulated mid-circuit measurements to verify projection based state preparation circuits. In this study, we test these methods across a variety of high-performance computing systems and our results show that circuits up to 21 qubits and more than 115,000,000 gates can be efficiently simulated.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Memory and Spin Injection Devices Involving Half Metals

We suggest memory and spin injection devices fabricated with half-metallic materials and based on the anomalous Hall effect. Schematic diagrams of the memory chips, in thin film and bulk crystal form, are presented. Spin injection devices made in thin film form are also suggested. These devices do not need any external magnetic field but make use of their own magnetization. Only a gate voltage is needed. The carriers are 100% spin polarized. Memory devices may potentially be smaller, faster, and less volatile than existing ones, and the injection devices may be much smaller and more efficient than existing spin injection devices.

Shaughnessy, M.↗

High-performance epistasis detection in quantitative trait GWAS

epiSNP is a program for identifying pairwise single nucleotide polymorphism (SNP) interactions (epistasis) in quantitative-trait genome-wide association studies (GWAS). A parallel MPI version (EPISNPmpi) was created in 2008 to address this computationally expensive analysis on large data sets with many quantitative traits and SNP markers. However, the falling cost of genotyping has led to an explosion of large-scale GWAS data sets that challenge EPISNPmpi’s ability to compute results in a reasonable amount of time. Therefore, we optimized epiSNP for modern multi-core and highly parallel many-core processors to efficiently handle these large data sets. This paper describes the serial optimizations, dynamic load balancing using MPI-3 RMA operations, and shared-memory parallelization with OpenMP to further enhance load balancing and allow execution on the Intel Xeon Phi coprocessor (MIC). For a large GWAS data set, our optimizations provided a 38.43× speedup over EPISNPmpi on 126 nodes using 2 MICs on TACC’s Stampede Supercomputer. We also describe a Coarray Fortran (CAF) version that demonstrates the suitability of PGAS languages for problems with this computational pattern. We show that the Coarray version performs competitively with the MPI version on the NERSC Edison Cray XC30 supercomputer. Finally, the performance benefits of hyper-threading for this application on Edison (average 1.35× speedup) are demonstrated.

97 MATHEMATICS AND COMPUTING↗

Minimum wafer thickness by rotated ingot ID wafering

The efficient utilization of materials is critical to certain device applications such as silicon for photovoltaics or diodes and gallium-gadolinium-garnet for memories. A variety of slicing techniques has been investigated to minimize wafer thickness and wafer kerf. This paper presents the results of analyses of ID wafering of rotated ingots based on predicted fracture behavior of the wafer as a result of forces during wafering and the properties of the device material. The analytical model indicated that the minimum wafer thickness is controlled by the depth of surface damage and the applied cantilever force. Both of these factors should be minimized. For silicon, a minimum thickness was found to be approximately 200 x 10 - 6th m for conventional sizes of rotated ingot wafering. Fractures through the thickness of the wafer rather than through the center supporting column were found to limit the minimum wafer thickness. The model suggested that the use of a vacuum chuck on the wafer surface to enhance cleavage fracture of the center supporting core and, with silicon, by using 111-line-type ingots could have potential for reducing minimum wafer thickness.

Chen, C. P.↗

Performance measurements of a data link protocol

The end-to-end performance of a program implementing IEEE Std 802.2 logical link control protocol, which provides virtual circuit and datagram service and which constitutes the top sublayer of the data link layer of the reference model for Local Area Networks is studied. The measurements of relative execution times of various functions required in transferring data, over virtual circuits or as datagrams, showed that most of the program execution time is spent in a few relatively low-level functions. One of these functions performs block movement of data in memory, and two other functions provide access to various queues. Based on this result, two alternative versions of the program which move data efficiently were written and measured. In this paper the results of the study are reported and their impact on designing hardware and software systems for a network interface are discussed.

Kanakia, Hemant↗

Heterogeneous concurrent computing with exportable services

Heterogeneous concurrent computing, based on the traditional process-oriented model, is approaching its functionality and performance limits. An alternative paradigm, based on the concept of services, supporting data driven computation, and built on a lightweight process infrastructure, is proposed to enhance the functional capabilities and the operational efficiency of heterogeneous network-based concurrent computing. TPVM is an experimental prototype system supporting exportable services, thread-based computation, and remote memory operations that is built as an extension of and an enhancement to the PVM concurrent computing system. TPVM offers a significantly different computing paradigm for network-based computing, while maintaining a close resemblance to the conventional PVM model in the interest of compatibility and ease of transition Preliminary experiences have demonstrated that the TPVM framework presents a natural yet powerful concurrent programming interface, while being capable of delivering performance improvements of upto thirty percent.

Sunderam, Vaidy↗

Ellicott City Disasters II: Enhancing a Statistical Flood Risk Model to Continue Improving Early Warning Systems and Public Safety in Ellicott City, Maryland

As flooding events in the United States grow in frequency and intensity, the use of technological advancements and applied science are increasingly necessary for effective flood monitoring and warning systems. The NASA DEVELOP Ellicott City Disasters II project investigated the use of machine learning for applications in flood risk detection to support the improvement of early warning systems. To strengthen the efforts of the Howard County Office of Emergency Management (OEM) in building a more robust flood monitoring system, the project improved the original statistical flood risk model, FLuME (Flood Learning Model Environment), programmed by the first DEVELOP term. The enhancements incorporated an additional six years of precipitation and soil moisture data from the North American Land Data Assimilation System (NLDAS), modeled using Aqua Advanced Microwave Scanning Radiometer for EOS and Tropical Rainfall Measuring Mission (TRMM) Microwave Imager. These Earth observations were supplemented by stream gauge data from the OEM and the US Geological Survey. The resultant flood risk model FLASH (Flood Learning Environment and Severity Assessment Hub) was trained to evaluate input variables and predict stage height in Ellicott City in real time. The addition of an advanced deep learning framework known as long short-term memory improved the model’s ability to capture relationships between variables. To assess the effectiveness of the new model, FLASH produced a model efficiency metric of 0.99, a significant improvement over the 0.85 value produced by the previous model. The project assisted the OEM in pursuing the integration of open data and NASA Earth observations into a threat matrix capable of informing near real-time decision making.

Disasters↗

Ellicott City Disasters II: Enhancing a Statistical Flood Risk Model to Continue Improving Early Warning Systems and Public Safety in Ellicott City, Maryland

As flooding events in the United States grow in frequency and intensity, the use of technological advancements and applied science are increasingly necessary for effective flood monitoring and warning systems. The NASA DEVELOP Ellicott City Disasters II project investigated the use of machine learning for applications in flood risk detection to support the improvement of early warning systems. To strengthen the efforts of the Howard County Office of Emergency Management (OEM) in building a more robust flood monitoring system, the project improved the original statistical flood risk model, FLuME (Flood Learning Model Environment), programmed by the first DEVELOP term The enhancements incorporated an additional six years of precipitation and soil moisture data from the North American Land Data Assimilation System (NLDAS), modeled using Aqua Advanced Microwave Scanning Radiometer for EOS and Tropical Rainfall Measuring Mission TRMM Microwave Imager. These Earth observations were supplemented by stream gauge data from the OEM and the US Geological Survey. The resultant flood risk model FLASH (Flood Learning Environment and Severity Assessment Hub) was trained to evaluate input variables and predict stage height in Ellicott City in real time. The addition of an advanced deep learning framework known as long short-term memory improved the model’s ability to capture relationships between variables. To assess the effectiveness of the new model, FLASH produced a model efficiency metric of 0.99, a significant improvement over the 0.85 value produced by the previous model. The project assisted the OEM in pursuing the integration of open data and NASA Earth observations into a threat matrix capable of informing near real-time decision making.

Disasters↗

Machine learning assisted hybrid models can improve streamflow simulation in diverse catchments across the conterminous US

Incomplete representations of physical processes often lead to structural errors in process-based (PB) hydrologic models. Machine learning (ML) algorithms can reduce streamflow modeling errors but do not enforce physical consistency. As a result, ML algorithms may be unreliable if used to provide future hydroclimate projections where climates and land use patterns are outside the range of training data. Here we test hybrid models built by integrating PB model outputs with a ML algorithm known as Long Short-Term Memory (LSTM) network on their ability to simulate streamflow in 531 catchments representing diverse conditions across the Conterminous United States. Model performance of hybrid models as measured by Nash-Sutcliffe efficiency (NSE) improved relative to standalone PB and LSTM models. More importantly, hybrid models provide highest improvement in catchments where PB models fail completely (i.e., NSE < 0). However, all models performed poorly in catchments with extended low flow periods, suggesting need for additional research.

54 ENVIRONMENTAL SCIENCES↗

Software Defined Architectures for Data Analytics

Data analytics applications increasingly are complex workflows composed of phases with very different program behaviors (e.g., graph algorithms and machine learning, algorithms operating on sparse an dense data structures, etc). To reach the levels of efficiency required to process these workflows in real time, upcoming architectures will need to leverage even more workload specialization. If, at one end, we may find even more heterogenous processors composed by a myriad of specialized processing elements, at the other end we may see novel reconfigurable architectures, composed of sets of functional units and memories interconnected with (re)configurable on-chip networks, able to adapt dynamically to adapt the workload characteristics. Field Programmable Gate Arrays are more and more used for accelerating various workloads and, in particular, inferencing in machine learning, providing higher efficiency than other solutions. However, their fine grained nature still leads to issues for the design software and still makes dynamic reconfiguration impractical. Future, more coarse-grained architectures could offer the features to execute diverse workloads at high efficiency while providing better reconfiguration mechanisms for dynamic adaptability. Nevertheless, we argue that the challenges for reconfigurable computing remain in the software. In this position paper, we describe a possible toolchain for reconfigurable architectures targeted at data analytics.

FPGA, Dataflow, Reconfigurable Architectures, Soft↗

Sage: parallel semi-asymmtric graph algorithms for NVRAMs

Non-volatile main memory (NVRAM) technologies provide an attractive set of features for large-scale graph analytics, including byte-addressability, low idle power, and improved memory-density. NVRAM systems today have an order of magnitude more NVRAM than traditional memory (DRAM). NVRAM systems could therefore potentially allow very large graph problems to be solved on a single machine, at a modest cost. However, a significant challenge in achieving high performance is in accounting for the fact that NVRAM writes can be much more expensive than NVRAM reads. In this paper, we propose an approach to parallel graph analytics using the Parallel Semi-Asymmetric Model (PSAM), in which the graph is stored as a read-only data structure (in NVRAM), and the amount of mutable memory is kept proportional to the number of vertices. Similar to the popular semi-external and semi-streaming models for graph analytics, the PSAM approach assumes that the vertices of the graph fit in a fast read-write memory (DRAM), but the edges do not. In NVRAM systems, our approach eliminates writes to the NVRAM, among other benefits. To experimentally study this new setting, we develop Sage, a parallel semi-asymmetric graph engine with which we implement provably-efficient (and often work-optimal) PSAM algorithms for over a dozen fundamental graph problems. We experimentally study Sage using a 48--core machine on the largest publicly-available real-world graph (the Hyperlink Web graph with over 3.5 billion vertices and 128 billion edges) equipped with Optane DC Persistent Memory, and show that Sage outperforms the fastest prior systems designed for NVRAM. Importantly, we also show that Sage nearly matches the fastest prior systems running solely in DRAM, by effectively hiding the costs of repeatedly accessing NVRAM versus DRAM.

97 MATHEMATICS AND COMPUTING↗