Recent Graphics Processing Units Research Articles

Wind Tunnel and Grid Resolution Effects in Large-Eddy Simulations of the High-Lift Common Research Model

This paper describes the application of a compressible large-eddy simulation (LES) solver to the flow over the High-Lift Common Research Model (CRM-HL) in landing configuration, a complex external aerodynamic flow configuration with deployed slats, flaps, a flow-through nacelle, and the associated brackets/fairings on the high-lift devices. The bulk Mach number is (0.2) and the mean-aerodynamic-chord-based Reynolds number is typical of a wind tunnel experiment (5.49E6). A key development in these simulations relative to previous work is that this work serves to establish robustness of LES methods to Reynolds number and aircraft configuration. Previous studies focused on the Japan Aerospace Exploration Agency Standard Model, which featured a nearly 3× lower Reynolds number than the present CRM-HL investigation and less aggressive wing/slat leading-edge curvature than the CRM-HL geometry, which led to lower leading-edge pressure suction peak magnitudes, both of which led to less stringent grid resolution requirements. In these LES simulations, an algebraic equilibrium wall modeling approach is employed along with a dynamic implementation of the Smagorinsky subgrid-scale model. The calculations are carried out in both a free air setting and one that includes the wind tunnel facility at seven angles of attack at five grid resolution levels, ranging from ≈10 to 1500 million control volumes. In free air, the solutions show decreasing sensitivity to the grid with each successive refinement level and systematically approach the experimental lift coefficient data as the grid is refined, with the 1.5 billion control volume case showing excellent agreement with the corrected experimental data. The simulations in both free air and in the wind tunnel predict a stall mechanism featuring a large inboard juncture stall and a nose-down break in the pitching moment curve, both of which agree with the experimental observations. The accuracy of the simulations is assessed via comparisons of integrated forces/moments, surface pressures, and surface skin friction visualizations. Graphics processing unit (GPU) acceleration of the charLES solver results in tractable turnaround times that make LES a useful tool in the aerospace industry design cycle. Recent GPU acceleration of the flow solver has made LES solutions that are highly accurate in lift/drag/moment for relevant high-lift aircraft flows achievable within about 5 h of wall time on 600 GPU cores.

Accelerating Open Modification Spectral Library Searching on Tensor Core in High-dimensional Space.

Driven by technological advances, the throughput and cost of mass spectrometry proteomics experiments have improved by orders of magnitude in recent decades. Spectral library searching is a common approach to annotating experimental mass spectra by matching them against large libraries of reference spectra corresponding to known peptides. An important disadvantage, however, is that only peptides included in the spectral library can be found, whereas novel peptides, such as those with unexpected post-translational modifications, will remain unknown. Open modification searching is an increasingly popular approach to annotate modified peptides based on partial matches against their unmodified counterparts. Unfortunately, this leads to very large search spaces and excessive runtimes, which is especially problematic considering the continuously increasing sizes of mass spectrometry proteomics datasets. We propose an open modification searching algorithm, called HOMS-TC, that fully exploits parallelism in the entire pipeline of spectral library searching. We designed a new highly parallel encoding method based on the principle of hyperdimensional computing to encode mass spectral data to hypervectors while minimizing information loss. This process can be easily parallelized since each dimension is calculated independently. HOMS-TC processes two stages of existing cascade search in parallel and selects the most similar spectra while considering post-translational modifications. We accelerate HOMS-TC on NVIDIA's Tensor Core Units, which is emerging and readily available in the recent graphics processing unit (GPU). Our evaluation shows that HOMS-TC is 31 × faster on average than alternative search engines and provides comparable accuracy to competing search tools. HOMS-TC is freely available under the Apache 2.0 license as an open-source software project at https://github.com/tycheyoung/homs-tc.

Development of a hardware-accelerated simulation kernel for ultra-high vacuum with Nvidia RTX GPUs

Molflow+ is a Monte Carlo (MC) simulation software for ultra-high vacuum, mainly used to simulate pressure in particle accelerators. In this article, we present and discuss the design choices arising in a new implementation of its ray-tracing–based simulation unit for Nvidia RTX Graphics Processing Units (GPUs). The GPU simulation kernel was designed with Nvidia’s OptiX 7 API to make use of modern hardware-accelerated ray-tracing units, found in recent RTX series GPUs based on the Turing and Ampere architectures. Even with the challenges posed by switching to 32 bit computations, our kernel runs much faster than on comparable CPUs at the expense of a marginal drop in calculation precision.

The Tensor-Core Correlator

Correlators are key components of radio telescopes as they combine the data from all receivers. They are characterized by high data rates, impressive compute loads, and the necessity to process data in real time. Often, these instruments use graphics processing units (GPUs) to correlate the incoming data streams as GPUs are fast, energy efficient, flexible, and programmable with a reasonable programming effort. Recent GPUs are equipped with a new technology, called “tensor cores”, to perform specific (typically artificial intelligence) workloads an order of magnitude more efficiently than regular GPU cores. This paper introduces the Tensor-Core Correlator, a GPU library that exploits this technology for signal processing and allows a GPU to correlate signals five to ten times faster and more energy efficiently than traditional state-of-the-art GPU correlators. The library hides the complexity of the use of tensor cores and can be easily integrated into the GPU pipelines of existing and future instruments, leading to a significant reduction in costs and energy consumption.

Architecture exploration of recent GPUs to analyze the efficiency of hardware resources

This study analyzes the efficiency of parallel computational applications with the adoption of recent graphics processing units (GPUs). We investigate the impacts of the additional resources of recent architecture on the popular benchmarks compared with previous architecture. Our simulation results demonstrate that Pascal GPU architecture improves the performance by 273% on average compared to old-fashioned Fermi architecture. To evaluate the performance improvement depending on specific hardware resources, we divide the hardware resources into two types: computing and memory resources. Computing resources have bigger impact on performance improvement than memory resources in most of benchmarks. For Hotspot and B+ tree, the architecture adopting only enhanced computing resources can achieve similar performance gains of the architecture adopting both computing and memory resources. We also evaluate the influence of the number of warp schedulers in the SM (Streaming Multiprocessor) to the GPU performance in relationship with barrier waiting time. Based on these analyses, we propose the development direction for the future generation of GPUs.

New Two-Level L1 Data Cache Bypassing Technique for High Performance GPUs

On-chip caches of graphics processing units (GPUs) have contributed to improved GPU performance by reducing long memory access latency. However, cache efficiency remains low despite the facts that recent GPUs have considerably mitigated the bottleneck problem of L1 data cache. Although the cache miss rate is a reasonable metric for cache efficiency, it is not necessarily proportional to GPU performance. In this study, we introduce a second key determinant to overcome the problem of predicting the performance gains from L1 data cache based on the assumption that miss rate only is not accurate. The proposed technique estimates the benefits of the cache by measuring the balance between cache efficiency and throughput. The throughput of the cache is predicted based on the warp occupancy information in the warp pool. Then, the warp occupancy is used for a second bypass phase when workloads show an ambiguous miss rate. In our proposed architecture, the L1 data cache is turned off for a long period when the warp occupancy is not high. Our two-level bypassing technique can be applied to recent GPU models and improves the performance by 6% on average compared to the architecture without bypassing. Moreover, it outperforms the conventional bottleneck-based bypassing techniques.

Modular Microservice based GPU Utilization Manager with Gunicorn

:Graphics processing unit (GPU) is a computer programmable chip that could perform rapid mathematical operations that can be accelerated with massive parallelism. In the early days, central processing unit (CPU) was responsible for all computations irrespective of whether it is feasible for parallel computation. However, in recent years GPUs are increasingly used for massively parallel computing applications, such as training Deep Neural Networks. GPU’s performance monitoring plays a key role in this new era since GPUs serve an inevitable role in increasing the speed of analysis of the developed system. GPU administration comes in picture to efficiently utilize the GPU when we deal with multiple workloads to run on the same hardware. In this study, various GPUparameters are monitored and help to keep them in safe levels and also to keep the improved performance of the system. This study,

Seismic modeling and inversion using half-precision floating-point numbers

New processors are increasingly supporting half-precision floating-point numbers, often with a significant throughput gain over single-precision operations. Seismic modeling, imaging, and inversion could benefit from such an acceleration, but it is not obvious how the accuracy of the solution can be preserved with a very narrow 16-bit representation. By scaling the finite-difference expression of the isotropic elastic wave equation, we have found that a stable solution can be obtained despite the very narrow dynamic range of the half-precision format.We develop an implementation with the CUDA platform, which, on most recent graphics processing units (GPU), is nearly twice as fast and uses half the memory of the equivalent single-precision version. The error on seismograms caused by the reduced precision is shown to correspond to a negligible fraction of the total seismic energy and is mostly incoherent with seismic phases. Finally, we find that this noise does not adversely impact full-waveform inversion nor reverse time migration, which both benefit from the higher throughput of half-precision computation.

Metric Selection for GPU Kernel Classification

Graphics Processing Units (GPUs) are vastly used for running massively parallel programs. GPU kernels exhibit different behavior at runtime and can usually be classified in a simple form as either “compute-bound” or “memory-bound.” Recent GPUs are capable of concurrently running multiple kernels, which raises the question of how to most appropriately schedule kernels to achieve higher performance. In particular, co-scheduling of compute-bound and memory-bound kernels seems promising. However, its benefits as well as drawbacks must be determined along with which kernels should be selected for a concurrent execution. Classifying kernels can be performed online by instrumentation based on performance counters. This work conducts a thorough analysis of the metrics collected from various benchmarks from Rodinia and CUDA SDK. The goal is to find the minimum number of effective metrics that enables online classification of kernels with a low overhead. This study employs a wrapper-based feature selection method based on the Fisher feature selection criterion. The results of experiments show that to classify kernels with a high accuracy, only three and five metrics are sufficient on a Kepler and a Pascal GPU, respectively. The proposed method is then utilized for a runtime scheduler. The results show an average speedup of 1.18× and 1.1× compared with a serial and a random scheduler, respectively.

GPGPU Virtualization Techniques a Comparative Survey

The Graphic Processing Units (GPU) are being adopted in many High Processing Computing (HPC) facilities because of their massively parallel and extraordinary computing power, which makes it possible to accelerate many general purpose implementations from different domains. A general-purpose GPU (GPGPU) is a GPU that performs computations that were traditionally handled by central processing unit (CPU) to accelerate applications along with handling traditional computations for graphics rendering. However, GPUs have some limitations, such as increased acquisition costs as well as larger space requirements, more powerful energy supplies, and their utilization is usually low for most workloads. That results in the need of GPU virtualization to maximize the use of an acquired GPU to share between the virtual machines for optimal use, to reduce power utilization and minimize the costs. This study comparatively reviews the recent GPU virtualization techniques including API remoting, para, full and hardware based virtualization, targeted for general-purpose accelerations.

Application Characteristics-Aware Sporadic Cache Bypassing for high performance GPGPUs

Modern graphics processing units (GPUs) with massive parallel architecture can boost the performance of both graphics and general-purpose applications. With the support of new programming tools, GPUs have become one of the most attractive platforms in the exploitation of the high thread-level parallelism. In the recent GPUs, hierarchical cache memories have been employed to support irregular memory-access patterns. However, the L1 data cache exhibits a poor efficiency in GPUs, and this is mainly due to the cache contention and the resource congestion. This paper shows that the L1 data cache does not always positively impact applications in terms of the performance; in fact, many applications are even slowed down due to the use of the L1 data cache. In this paper, a novel cache bypassing mechanism (CARB) is proposed to increase the efficiency of the GPU cache management and to improve the GPU performance. The CARB mechanism exploits the characteristics of the currently executed applications to estimate the performance impact of the L1 data cache on the GPU, and it then allows memory requests to bypass the cache in discrete phases during the execution time. The bypassing decision is determined adaptively at runtime. Experiment results show that the CARB mechanism achieves an average speedup of 22% for a wide range of GPGPU applications.

Investigation of the flow structure in thin polymer films using 3D µPTV enhanced by GPU

To understand the effects of inhomogeneous drying on the quality of polymer coatings, an experimental setup to resolve the occurring flow field throughout the drying film has been developed. Deconvolution microscopy is used to analyze the flow field in 3D and time. Since the dimension of the spatial component in the direction of the line-of-sight is limited compared to the lateral components, a multi-focal approach is used. Here, the beam of light is equally distributed on up to five cameras using cubic beam splitters. Adding a meniscus lens between each pair of camera and beam splitter and setting different distances between each camera and its meniscus lens creates multi-focality and allows one to increase the depth of the observed volume. Resolving the spatial component in the line-of-sight direction is based on analyzing the point spread function. The analysis of the PSF is computational expensive and introduces a high complexity compared to traditional particle image velocimetry approaches. A new algorithm tailored to the parallel computing architecture of recent graphics processing units has been developed. The algorithm is able to process typical images in less than a second and has further potential to realize online analysis in the future. As a prove of principle, the flow fields occurring in thin polymer solutions drying at ambient conditions and at boundary conditions that force inhomogeneous drying are presented.

Papaya Tree Detection with UAV Images Using a GPU-Accelerated Scale-Space Filtering Method

The use of unmanned aerial vehicles (UAV) can allow individual tree detection for forest inventories in a cost-effective way. The scale-space filtering (SSF) algorithm is commonly used and has the capability of detecting trees of different crown sizes. In this study, we made two improvements with regard to the existing method and implementations. First, we incorporated SSF with a Lab color transformation to reduce over-detection problems associated with the original luminance image. Second, we ported four of the most time-consuming processes to the graphics processing unit (GPU) to improve computational efficiency. The proposed method was implemented using PyCUDA, which enabled access to NVIDIA’s compute unified device architecture (CUDA) through high-level scripting of the Python language. Our experiments were conducted using two images captured by the DJI Phantom 3 Professional and a most recent NVIDIA GPU GTX1080. The resulting accuracy was high, with an F-measure larger than 0.94. The speedup achieved by our parallel implementation was 44.77 and 28.54 for the first and second test image, respectively. For each 4000 × 3000 image, the total runtime was less than 1 s, which was sufficient for real-time performance and interactive application.

Efficient Kernel Management on GPUs

Graphics Processing Units (GPUs) have been widely adopted as accelerators for compute-intensive applications due to its tremendous computational power and high memory bandwidth. As the complexity of applications continues to grow, each new generation of GPUs has been equipped with advanced architectural features and more resources to sustain its performance acceleration capability. Recent GPUs have been featured with concurrent kernel execution, which is designed to improve the resource utilization by executing multiple kernels simultaneously. However, it is still a challenge to find a way to manage the resources on GPUs for concurrent kernel execution. Prior works only achieve limited performance improvement as they do not optimize the thread-level parallelism (TLP) and model the resource contention for the concurrently executing kernels. In this article, we design an efficient kernel management framework that optimizes the performance for concurrent kernel execution on GPUs. Our kernel management framework contains two key components: TLP modulation and cache bypassing. The TLP modulation is employed to adjust the TLP for the concurrently executing kernels. It consists of three parts: kernel categorization, static TLP modulation, and dynamic TLP modulation. The cache bypassing is proposed to mitigate the cache contention by only allowing a subset of a kernel’s blocks to access the L1 data cache. Experiments indicate that our framework can improve the performance by 1.51 × on average (energy-efficiency by 1.39 × on average), compared with the default concurrent kernel execution framework.

A Fast Discrete Wavelet Transform Using Hybrid Parallelism on GPUs

Wavelet transform has been widely used in many signal and image processing applications. Due to its wide adoption for time-critical applications, such as streaming and real-time signal processing, many acceleration techniques were developed during the past decade. Recently, the graphics processing unit (GPU) has gained much attention for accelerating computationally-intensive problems and many solutions of GPU-based discrete wavelet transform (DWT) have been introduced, but most of them did not fully leverage the potential of the GPU. In this paper, we present various state-of-the-art GPU optimization strategies in DWT implementation, such as leveraging shared memory, registers, warp shuffling instructions, and thread- and instruction-level parallelism (TLP, ILP), and finally elaborate our hybrid approach to further boost up its performance. In addition, we introduce a novel mixed-band memory layout for Haar DWT, where multi-level transform can be carried out in a single fused kernel launch. As a result, unlike recent GPU DWT methods that focus mainly on maximizing ILP, we show that the optimal GPU DWT performance can be achieved by hybrid parallelism combining both TLP and ILP together in a mixed-band approach. We demonstrate the performance of our proposed method by comparison with other CPU and GPU DWT methods.

GPU-based RFA simulation for minimally invasive cancer treatment of liver tumours.

Radiofrequency ablation (RFA) is one of the most popular and well-standardized minimally invasive cancer treatments (MICT) for liver tumours, employed where surgical resection has been contraindicated. Less-experienced interventional radiologists (IRs) require an appropriate planning tool for the treatment to help avoid incomplete treatment and so reduce the tumour recurrence risk. Although a few tools are available to predict the ablation lesion geometry, the process is computationally expensive. Also, in our implementation, a few patient-specific parameters are used to improve the accuracy of the lesion prediction. Advanced heterogeneous computing using personal computers, incorporating the graphics processing unit (GPU) and the central processing unit (CPU), is proposed to predict the ablation lesion geometry. The most recent GPU technology is used to accelerate the finite element approximation of Penne's bioheat equation and a three state cell model. Patient-specific input parameters are used in the bioheat model to improve accuracy of the predicted lesion. A fast GPU-based RFA solver is developed to predict the lesion by doing most of the computational tasks in the GPU, while reserving the CPU for concurrent tasks such as lesion extraction based on the heat deposition at each finite element node. The solver takes less than 3 min for a treatment duration of 26 min. When the model receives patient-specific input parameters, the deviation between real and predicted lesion is below 3 mm. A multi-centre retrospective study indicates that the fast RFA solver is capable of providing the IR with the predicted lesion in the short time period before the intervention begins when the patient has been clinically prepared for the treatment.

A Visualization System for Interactive Exploration of the Cardiac Anatomy.

Because of the complex and fine structure, visualization of the heart still remains a challenging task, which makes it an active research topic. In this paper, we present a visualization system for medical data, which takes advantage of the recent graphics processing unit (GPU) and can provide real-time cardiac visualization. This work focuses on investigating the anatomical structure visualization of the human heart, which is fundamental to the cardiac visualization, medical training and diagnosis assistance. Several state-of-the-art cardiac visualization methods are integrated into the proposed system and a task specified visualization method is proposed. In addition, auxiliary tools are provided to generate user specified visualization results. The contributions of our work lie in two-fold: for doctors and medical staff, the system can provide task specified visualization with interactive visualization tools; for researchers, the proposed platform can serve as a baseline for comparing different rendering methods and can easily incorporate new rendering methods. Experimental results show that the proposed system can provide favorable cardiac visualization results in terms of both effectiveness and efficiency.

Comparison of Acceleration Data Structures for Electromagnetic Ray-Tracing Purposes on GPUs [EM Programmer's Notebook

We analyze and compare the performances of two acceleration data structures for electromagnetic ray-tracing purposes on graphical processing units (GPUs) using the CUDA programming language, namely the K-Dimensional (KD)-tree and the Split Bounding Volume Hierarchy (SBVH). Our implementations have been based on the approach made available by Nvidia, which takes into account the programming optimizations made possible by the latest version of CUDA and the most recent Nvidia GPU architectures. We have tested the two approaches on standard computer graphics scenes (conference and bunny) and on a scene of electromagnetic interest (ship). In all the cases considered, the SBVH has shown to perform better in terms of both speed and memorysaving properties.

Thermal link-wise artificial compressibility method: GPU implementation and validation of a double-population model

The link-wise artificial compressibility method (LW-ACM) is a novel formulation of the artificial compressibility method for the incompressible Navier–Stokes equations showing strong analogies with the lattice Boltzmann method (LBM). The LW-ACM operates on regular Cartesian meshes and is therefore well-suited for massively parallel processors such as graphics processing units (GPUs). In this work, we describe the GPU implementation of a three-dimensional thermal flow solver based on a double-population LW-ACM model. Focusing on large scale simulations of the differentially heated cubic cavity, we compare the present method to hybrid approaches based on either multiple-relaxation-time LBM (MRT-LBM) or LW-ACM, where the energy equation is solved through finite differences on a compact stencil. Since thermal LW-ACM requires only the storing of fluid density and velocity in addition to temperature, both double-population thermal LW-ACM and hybrid thermal LW-ACM reduce the memory requirements by a factor of 4.4 compared to a D3Q19 hybrid thermal LBM implementation following a two-grid approach. Using a single graphics card featuring 6GiB11Instead of the widespread but ambiguous GB and KB notations, we use the notations of the International System of Quantities, namely 1GiB=230B, 1KiB=210B, and 1kB=103B. of memory, we were able to perform single-precision computations on meshes containing up to 5363 nodes, i.e. about 154 million nodes. We show that all three methods are comparable both in terms of accuracy and performance on recent GPUs. For Rayleigh numbers ranging from 104 to 106, the thermal fluxes as well as the flow features are in similar good agreement with reference values from the literature.

Directive-based GPU programming for computational fluid dynamics

Directive-based programming of graphics processing units (GPUs) has recently appeared as a viable alternative to using specialized low-level languages such as CUDA C and OpenCL for general-purpose GPU programming. This technique, which uses “directive” or “pragma” statements to annotate source codes written in traditional high-level languages, is designed to permit a unified code base to serve multiple computational platforms. In this work we analyze the popular OpenACC programming standard, as implemented by the PGI compiler suite, in order to evaluate its utility and performance potential in computational fluid dynamics (CFD) applications. We examine the process of applying the OpenACC Fortran API to a test CFD code that serves as a proxy for a full-scale research code developed at Virginia Tech; this test code is used to asses the performance improvements attainable for our CFD algorithm on common GPU platforms, as well as to determine the modifications that must be made to the original source code in order to run efficiently on the GPU. Performance is measured on several recent GPU architectures from NVIDIA and AMD (using both double and single precision arithmetic) and the accelerator code is benchmarked against a multithreaded CPU version constructed from the same Fortran source code using OpenMP directives. A single NVIDIA Kepler GPU card is found to perform approximately 20× faster than a single CPU core and more than 2× faster than a 16-core Xeon server. An analysis of optimization techniques for OpenACC reveals cases in which manual intervention by the programmer can improve accelerator performance by up to 30% over the default compiler heuristics, although these optimizations are relevant only for specific platforms. Additionally, the use of multiple accelerators with OpenACC is investigated, including an experimental high-level interface for multi-GPU programming that automates scheduling tasks across multiple devices. While the overall performance of the OpenACC code is found to be satisfactory, we also observe some significant limitations and restrictions imposed by the OpenACC API regarding certain useful features of modern Fortran (2003/8); these are sufficient for us to conclude that it would not be practical to apply OpenACC to our full research code at this time due to the amount of refactoring required.