Search NASA⌕ Search

SEARCH · Search NASA

Results for “hardware accelerators”

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 901 records · Page 50

Protection of Hardware: Powering Systems (Power Converter, Normal Conducting, and Superconducting Magnets)

Along with the protection of magnets and power converters, we have added a section on personnel protection because this is our highest priority in the design and operation of power systems. Thus, our topics are the protection of people, power converters, and magnet loads (protected from the powering equipment), including normal conducting magnets and superconducting magnets.

43 PARTICLE ACCELERATORS↗

ECP SOLLVE: Validation and Verification Testsuite Status Update and Compiler Insight for OpenMP

The OpenMP language continues to evolve with every new specification release, as does the need to validate and verify the new features that have been implemented by the different vendors. With the release of OpenMP 5.0 and OpenMP 5.1, new target offload and host-based features have been introduced to the programming model. While OpenMP continues to grow in maturity, there is an observable growth in the number of compiler and hardware vendors that support OpenMP. In this manuscript, the main focus is on evaluating the conformity and OpenMP implementation progress of various compiler vendors such as Cray, IBM, GNU, Clang/LLVM, NVIDIA, and Intel. More specifically, the 4.5, 5.0, and 5.1 versions of the OpenMP specification are analyzed. For our experimental setup, the Crusher and Summit computing systems hosted by Oak Ridge National Lab’s Computing Facilities are utilized. The effort of vendor agnostic analysis of these implementations is especially valuable for application developers who are using new OpenMP features to accelerate their scientific codes. Insights are presented into the current implementation status of various vendors, the progression of specific compiler’s support for OpenMP overtime, the subset of OpenMP 4.5, 5.0, and 5.1 that is supported by all compilers, and examples of how our test suite has influenced discussion regarding the correct interpretation of the OpenMP specification. By evaluating OpenMP conformity of pre-Exascale computing systems, the aim is to detail progress and status of AMD + Cray ecosystem before the system and their OpenMP implementation is used for mission critical applications when the first Exascale Computer Frontier is made available to applications.

Huber, Thomas↗

Performance of the LLRF System for the Fermilab PIP-II Injector Test

PIP-II IT is a test facility for the PIP-II project where the injector, warm front-end, and the first two superconducting cryomodules are being tested. The 8-cavity half-wave-resonator (HWR) cryomodule operating at 162.5 MHz is followed by the 8-cavity single-spoke resonator(SSR1) cryomodule operating at 325 MHz. The LLRF systems for both cryomodules are based on a common SOC FPGA-based hardware platform. The resonance control systems for the two cryomodules are quite different, the first being a pneumatic system based on helium pressure and the latter a piezo/stepper motor type control. The data acquisition and control system can support both CW and Pulsed mode operations. Beam loading compensation is available which can be used for both manual/automatic control in the LLRF system. The user interfaces include EPICS, Labview, and ACNET. Testing of the RF system has progressed to the point of being ready for a 2 mA beam to be accelerated to 25 MeV. The design and performance of the field control and resonance control system operation with beam are presented in this paper.

43 PARTICLE ACCELERATORS↗

Hls4ml Synthesis Testing

HLS4ml (high level synthesis for machine learning) Is a Python package used to translate commonly used open-source machine learning models into HLS. This is useful in machine learning applications on FPGAs. Machine learning algorithms are only as fast as the hardware that they are used on, and some applications require high speed without sacrificing accuracy. In these situations, an FPGA is a good choice since it is faster than a CPU or a GPU, but programming an FPGA is difficult. This is where HLS4ml can be used to simplify the process, as a well-known learning model can be converted to HLS and more easily deployed onto an FPGA. There are many use cases for a machine learning algorithm running on an FPGA. For example, detectors in a particle accelerator cannot keep every event that they detect, and so a computer must decide which events to keep and which to discard. Using an FPGA with a machine learning algorithm would be a good way to keep as many events as possible.

Swanson, Caiden↗

Final report LDRD project 105816 : model reduction of large dynamic systems with localized nonlinearities.

Advanced computing hardware and software written to exploit massively parallel architectures greatly facilitate the computation of extremely large problems. On the other hand, these tools, though enabling higher fidelity models, have often resulted in much longer run-times and turn-around-times in providing answers to engineering problems. The impediments include smaller elements and consequently smaller time steps, much larger systems of equations to solve, and the inclusion of nonlinearities that had been ignored in days when lower fidelity models were the norm. The research effort reported focuses on the accelerating the analysis process for structural dynamics though combinations of model reduction and mitigation of some factors that lead to over-meshing.

99 GENERAL AND MISCELLANEOUS↗

hls4ml

hls4ml (high level synthesis for machine learning) Is a Python package used to translate commonly used open-source machine learning models into HLS. This is useful in machine learning applications on FPGAs. Machine learning algorithms are only as fast as the hardware that they are used on, and some applications require high speed without sacrificing accuracy. In these situations, an FPGA is a good choice since it is faster than a CPU or a GPU, but programming an FPGA is difficult. This is where hls4ml can be used to simplify the process, as a well-known learning model can be converted to HLS and more easily deployed onto an FPGA. There are many use cases for a machine learning algorithm running on an FPGA. For example, detectors in a particle accelerator cannot keep every event that they detect, and so a computer must decide which events to keep and which to discard. Using an FPGA with a machine learning algorithm would be a good way to keep as many events as possible.

Swanson, Caiden↗

Soldered Contact and Current Risetime Effects on Negative Polarity Wire Array Z-pinches

The Cornell University COBRA pulser is a nominal 1 MA machine, capable of driving up to 32 wire cylindrical Z-pinch arrays. COBRA can operate with variable current risetimes ranging from 100 ns to 200 ns (short and long pulse, respectively). Wires are typically strung with a 'press' contact to the electrode hardware, where the wire is loosely pulled against the hardware and held there to establish electrical contact. The machine is normally negative, but a bolt-on convolute can be used to modify the current path and effectively produce positive polarity operation at the load.Previous research with single wires on a 1-5 kA pulser has shown that soldering the wire, thereby improving the wire/electrode contact, and operating in positive polarity can improve the energy deposition into the wire and enhance wire core expansion. Negative polarity showed no difference. Previous experiments on the negative polarity, 20 MA, 100 ns Z accelerator have shown that improving the contact improved the x-ray yield.Cornell data were collected on 16-wire Aluminum Z-pinch arrays in negative polarity. Experiments were conducted with both short and long current pulses with soldered and no-soldered wire/electrode contacts. The initiation, ablation, implosion and stagnation phases were compared for these four conditions. Time dependent x-ray signals were measured using diodes and diamond detectors. An inductive voltage monitor was used to infer minimum current radius achieved, as defined by a uniform shell of current moving radially inward, producing a time dependent inductance. Total energy data were collected with a metal-strip bolometer. Self-emission data were collected by an XUV 4-frame camera and an optical streak camera.In negative polarity and with short pulses, soldering appeared to produce a smaller radius pinch and decrease variations in the x-ray pulse shape. The bolometer, laser backlighter, 4-frame and streak cameras showed negligible differences in the initiation ablation and implosion phases. X-ray yield was increased with soldered arrays in negative polarity with long pulses. The bolometer also showed a 50% increase in radial power emitted for soldered arrays. 4-frame images showed soldered arrays have a more pronounced 'Christmas Tree' effect originating from the cathode. The inductive voltage monitor showed, with both long and short pulses, that soldered and no-solder arrays reached the same minimum current radius at the same time. However, with long pulses soldered arrays radiate x-rays at the time of minimum current radius while no-solder array x-ray output is delayed by {approx}20 ns, resulting in decreased x-ray yield.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

STAR Vertex Detector Upgrade-HFT Pixel Development

Development and prototyping efforts directed towards construction of a new vertex detector for the STAR experiment at the RHIC accelerator at BNL are presented. This new detector will extend the physics range of STAR by allowing for precision measurements of yields and spectra of particles containing heavy quarks. The innermost central part of the new detector is a high resolution pixel-type detector (PIXEL). PIXEL requirements are discussed as well as a conceptual mechanical design, a sensor development path, and a detector readout architecture. Selected progress with sensor prototypes dedicated to the PIXEL detector is summarized and the approach chosen for the readout system architecture validated in tests of hardware prototypes is discussed.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Fast IMRT with narrow high energy scanned photon beams

Purpose: Since the first publications on intensity modulated radiation therapy (IMRT) in the early 1980s almost all efforts have been focused on fairly time consuming dynamic or segmental multileaf collimation. With narrow fast scanned photon beams, the flexibility and accuracy in beam shaping increases, not least in combination with fast penumbra trimming multileaf collimators. Previously, experiments have been performed with full range targets, generating a broad bremsstrahlung beam, in combination with multileaf collimators or material compensators. In the present publication, the first measurements with fast narrow high energy (50 MV) scanned photon beams are presented indicating an interesting performance increase even though some of the hardware used were suboptimal. Methods: Inverse therapy planning was used to calculate optimal scanning patterns to generate dose distributions with interesting properties for fast IMRT. To fully utilize the dose distributional advantages with scanned beams, it is necessary to use narrow high energy beams from a thin bremsstrahlung target and a powerful purging magnet capable of deflecting the transmitted electron beam away from the generated photons onto a dedicated electron collector. During the present measurements the scanning system, purging magnet, and electron collimator in the treatment head of the MM50 racetrack accelerator was used with 3-6 mm thick bremsstrahlung targets of beryllium. The dose distributions were measured with diodes in water and with EDR2 film in PMMA. Monte Carlo simulations with geant4 were used to study the influence of the electrons transmitted through the target on the photon pencil beam kernel. Results: The full width at half-maximum (FWHM) of the scanned photon beam was 34 mm measured at isocenter, below 9.5 cm of water, 1 m from the 3 mm Be bremsstrahlung target. To generate a homogeneous dose distribution in a 10 x 10 cm{sup 2} field, the authors used a spot matrix of 100 equal intensity beam spots resulting in a uniformity of collimated 80%-20% penumbra of 9 mm at a primary electron energy of 50 MeV. For the more complex cardioid shaped dose distribution, they used 270 spots, which at a pulse repetition frequency of 200 Hz is completed every 1.36 s. Conclusions: The present measurements indicate that the use of narrow scanned photon beams is a flexible and fast method to deliver advanced intensity modulated beams. Fast scanned photon IMRT should, therefore, be a very interesting modality in the delivery of biologically optimized radiation therapy with the possibility for in vivo treatment verification with PET-CT imaging.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Fast jet tagging with MLP-Mixers on FPGAs

We explore the innovative use of MLP-Mixer models for real-time jet tagging and establish their feasibility on resource-constrained hardware like FPGAs. MLP-Mixers excel in processing sequences of jet constituents, achieving state-of-the-art performance on datasets mimicking Large Hadron Collider conditions. By using advanced optimization techniques such as High-Granularity Quantization and Distributed Arithmetic, we achieve unprecedented efficiency. These models match or surpass the accuracy of previous architectures, reduce hardware resource usage by up to 97%, double the throughput, and half the latency. Additionally, non-permutation-invariant architectures enable smart feature prioritization and efficient FPGA deployment, setting a new benchmark for machine learning in real-time data processing at particle colliders.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

An FPGA-based Pattern Recognition Associative Memory

Pattern recognition associative memory (PRAM) devices are parallel processing engines which are used to tackle the complex combinatorics of track nding algorithms, particularly for silicon based tracking triggers. PRAM development has been mostly limited to the realm of ASICs, which often leads to lengthy and expensive design cycles. FPGAs allow for quick iterations, making them an ideal hardware platform for designing and evaluating new PRAM features before committing to silicon. In this paper we present our FPGA-based PRAM designs and discuss how logic blocks which were originally developed for ASICs have been redesigned to take advantage of modern FPGA architectures.

43 PARTICLE ACCELERATORS↗

Development of Proton Computed Tomography for Applications in Proton Therapy

Determination of the Bragg peak position in proton therapy requires accurate knowledge of the electron density and ratio of effective atomic number and mass (Z/A) of the body tissues traversed. While the Z/A ratio is fairly constant for human tissues, the density of tissues varies significantly. One possibility to obtain accurate electron density information of tissues is to use protons of sufficient energy to penetrate the patient and measure their energy loss. From these transmission measurements, it is possible to reconstruct a three-dimensional map of electron densities using algebraic techniques. The interest in proton computed tomography (pCT) has considerably increased in recent years due to the more common use of proton accelerators for cancer treatment world-wide and a modern design concept based on current high-energy physics technology has been suggested. This contribution gives a status update on the pCT project carried out by the pCT Collaboration, a group of institutions sharing interest and expertise in the development of pCT. We will present updated imaging data obtained with a small pCT prototype developed in collaboration with the Santa Cruz Institute of Particle Physics and installed on the proton research beam line at Loma Linda University Medical Center. We will discuss hardware decisions regarding the next-generation pCT scanner, which will permit scanning of head-sized objects. Progress has also been made in the formulation of the most likely path of protons through an object and parallelizable iterative reconstruction algorithms that can be implemented on general-purpose commodity graphics processing units. Finally, we will present simulation studies for utilizing pCT technology for on-line proton dose verification and tumor imaging with positron emission tomography (PET)

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Randomized Algorithms for Scientific Computing (RASC)

Randomized algorithms have propelled advances in artificial intelligence (AI) and represent a foundational research area in advancing AI for Science. Future advancements in DOE Office of Science priority areas such as climate science, astrophysics, fusion, advanced materials, combustion, and quantum computing all require randomized algorithms for surmounting challenges of complexity, robustness, and scalability. Advances in data collection and numerical simulation have changed the dynamics of scientific research and motivate the need for randomized algorithms. For instance, advances in imaging technologies such as X-ray ptychography, electron microscopy, electron energy loss spectroscopy, or adaptive optics lattice light-sheet microscopy collect hyperspectral imaging and scattering data in terabytes, at breakneck speed enabled by state-of-the-art detectors. The data collection is exceptionally fast compared with its analysis. Likewise, advances in high-performance architectures have made exascale computing a reality and changed the economies of scientific computing in the process. Floating-point operations that create data are essentially free in comparison with data movement. Thus far, most approaches have focused on creating faster hardware. Ironically, this faster hardware has exacerbated the problem by making data still easier to create. Under such an onslaught, scientists often resort to heuristic deterministic sampling schemes (e.g., low-precision arithmetic, sampling every nth element) and sacrifice potentially valuable accuracy. Dramatically better results can be achieved via randomized algorithms, reducing the data size as much as or more than naive deterministic subsampling can achieve, while retaining the high accuracy of computing on the full data set. By randomized algorithms we mean those algorithms that employ some form of randomness in internal algorithmic decisions to accelerate time to solution, increase scalability, or improve reliability. Examples include matrix sketching for solving large-scale least-squares problems (see Figure 1) and stochastic gradient descent for training machine learning models. We are not recommending heuristic methods but rather randomized algorithms that have certificates of correctness and probabilistic guarantees of optimality and near-optimality. Such approaches can be useful beyond acceleration, for example, in understanding how to avoid measure zero worst-case scenarios that plague methods such as QR matrix factorization.

97 MATHEMATICS AND COMPUTING↗

Reliable edge machine learning hardware for scientific applications

Extreme data rate scientific experiments create massive amounts of data that require efficient ML edge processing. This leads to unique validation challenges for VLSI implementations of ML algorithms: enabling bit-accurate functional simulations for performance validation in experimental software frameworks, verifying those ML models are robust under extreme quantization and pruning, and enabling ultra-fine-grained model inspection for efficient fault tolerance. We discuss approaches to developing and validating reliable algorithms at the scientific edge under such strict latency, resource, power, and area requirements in extreme experimental environments. We study metrics for developing robust algorithms, present preliminary results and mitigation strategies, and conclude with an outlook of these and future directions of research towards the longer-term goal of developing autonomous scientific experimentation methods for accelerated scientific discovery.

Baldi, Tommaso↗

Synchronous detection of longitudinal and transverse bunch signals at a storage ring

To understand and control dynamics in the longitudinal phase space, time-resolved measurements of different bunch parameters are required. For a reconstruction of this phase space, the detector systems have to be synchronized. This reconstruction can be used for example for studies of the microbunching instability which occurs if the interaction of the bunch with its own radiation leads to the formation of substructures on the longitudinal bunch profile. These substructures can grow rapidly—leading to a sawtooth-like behavior of the bunch. At KARA, we use a fast-gated intensified camera for energy spread studies, Schottky diodes for coherent synchrotron radiation studies as well as electro-optical spectral decoding for longitudinal bunch profile measurements. For a synchronization, a synchronization scheme is used which compensates for hardware delays. Here in this paper, the different experimental setups and their synchronization are discussed and first results of synchronous measurements presented.

43 PARTICLE ACCELERATORS↗

A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU, In: 2014 IEEE 28th International Parallel and Distributed Processing Symposium

Power and energy consumption are becoming an increasing concern in high performance computing. Compared to multi-core CPUs, GPUs have a much better performance per watt. In this paper we discuss efforts to redesign the most computation intensive parts of BLAST, an application that solves the equations for compressible hydrodynamics with high order finite elements, using GPUs BLAST, Dobrev. In order to exploit the hardware parallelism of GPUs and achieve high performance, we implemented custom linear algebra kernels. We intensively optimized our CUDA kernels by exploiting the memory hierarchy, which exceed the vendor's library routines substantially in performance. We proposed an auto tuning technique to adapt our CUDA kernels to the orders of the finite element method. Compared to a previous base implementation, our redesign and optimization lowered the energy consumption of the GPU in two aspects: 60% less time to solution and 10% less power required. Compared to the CPU-only solution, our GPU accelerated BLAST obtained a 2.5× overall speedup and 1.42× energy efficiency (green up) using 4th order (Q_4) finite elements, and a 1.9× speedup and 1.27× green up using 2nd order (Q2) finite elements.

Computer Science↗

Directive-Based, High-Level Programming and Optimizations for High-Performance Computing with FPGAs

Reconfigurable architectures like Field Programmable Gate Arrays (FPGAs) have been used for accelerating computations from several domains because of their unique combination of flexibility, performance, and power efficiency. However, FPGAs have not been widely used for high-performance computing, primarily because of their programming complexity and difficulties in optimizing performance. In this paper, we present a directive-based, high-level optimization framework for high-performance computing with FPGAs, built on top of an OpenACC-to-FPGA translation framework called OpenARC. We propose directive extensions and correspond- ing compile-time optimization techniques to enable the compiler to generate more efficient FPGA hardware configuration files. Empirical evaluation of the proposed framework on an Intel Stratix V with five OpenACC benchmarks from various application domains shows that FPGA-specific optimizations can lead to significant increases in performance across all tested applications. We also demonstrate that applying these high-level directive-based optimizations can allow OpenACC applications to perform similarly to lower-level OpenCL applications with hand-written FPGA-specific optimizations, and offer runtime and power performance benefits compared to CPUs and GPUs.

Lambert, Jacob↗

A lightweight approach to performance portability with targetDP

Leading high performance computing systems achieve their status through use of highly parallel devices such as NVIDIA graphics processing units or Intel Xeon Phi many-core CPUs. The concept of performance portability across such architectures, as well as traditional CPUs, is vital for the application programmer. In this paper we describe targetDP, a lightweight abstraction layer which allows grid-based applications to target data parallel hardware in a platform agnostic manner. We demonstrate the effectiveness of our pragmatic approach by presenting performance results for a complex fluid application (with which the model was co-designed), plus separate lattice quantum chromodynamics particle physics code. For each application, a single source code base is seen to achieve portable performance, as assessed within the context of the Roofline model. TargetDP can be combined with Message Passing Interface (MPI) to allow use on systems containing multiple nodes: we demonstrate this through provision of scaling results on traditional and graphics processing unit-accelerated large scale supercomputers.

Computer Science↗