CUDA Research Articles

Dynamic GPU Scheduling With Multi-Resource Awareness and Live Migration Support

In clouds and data centers, GPU servers with multiple GPUs are widely deployed. Current state-of-the-art GPU scheduling policies are “static” in assigning applications to different GPUs. These policies usually ignore the dynamics of the GPU utilization and are often inaccurate in estimating resource demand before assigning/running applications, so there is a large opportunity to further balance the loads and improve GPU utilization. Based on CUDA (Compute Unified Device Architecture), we develop a runtime system called DCUDA which supports <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">“dynamic”</i> scheduling of running applications between multiple GPUs. In particular, DCUDA takes into consideration multidimensional resources, including computing cores, memory usage, and energy consumption. It first provides a real-time and lightweight method to accurately monitor the resource demand of applications and GPU utilization. Furthermore, it provides a universal migration facility to migrate <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">“running applications”</i> between GPUs with negligible overhead. More importantly, DCUDA transparently supports all CUDA applications without changing their source code. Experiments with our prototype system show that DCUDA can reduce 78.3% of overloaded time of GPUs on average. As a result, for different workloads consisting of a wide range of applications we studied, DCUDA can reduce the average execution time of general applications by up to 42.1%, and even up to 67% for memory-intensive tasks. Besides, DCUDA also reduces 13.3% of energy in light-load scenarios.

VOLUMETRIC LIGHT TRANSFER IN FUNCTIONALLY DEFINED SCENES

In computer graphics, the main goals are high realism and rendering speed. In terms of realism, physically correct rendering is important, which calculates the transmission of light from the source to the observer. A method of modeling global illumination is presented, which takes into account light coming directly from sources, as well as light redirected through other parts of the scene. This allows you to create various optical effects, such as the flow of color from one surface to another, indirect lighting and caustics. The photon display method is designed for functionally defined scenes using graphics processors, thereby achieving an interactive rendering mode. The aim of the work is to implement a method for displaying photons in interactive rendering mode using graphics processors. It was necessary to find a compromise between physical accuracy and the speed of displaying scenes. As the number of parallel computing cores increases, traditional programming models become less efficient. GPUs based on the parallel processing model are being expanded by adding improved support for general-purpose computing to their cores. CUDA is a multicore computing platform. GPU-enabled hardware provides low-cost access to many computations. The method presented in this paper extends the photon-mapping algorithm for the CUDA platform to volumetric photon mapping. The beam brightness estimate used to efficiently calculate the density is recovered with the correct physical parameters. Photons are traced and bandwidth is selected. Performance increases due to calculations on multiple cores and the use of large compacted blocks.

LSMMD-MA: scaling multimodal data integration for single-cell genomics data analysis.

Modality matching in single-cell omics data analysis-i.e. matching cells across datasets collected using different types of genomic assays-has become an important problem, because unifying perspectives across different technologies holds the promise of yielding biological and clinical discoveries. However, single-cell dataset sizes can now reach hundreds of thousands to millions of cells, which remain out of reach for most multimodal computational methods. We propose LSMMD-MA, a large-scale Python implementation of the MMD-MA method for multimodal data integration. In LSMMD-MA, we reformulate the MMD-MA optimization problem using linear algebra and solve it with KeOps, a CUDA framework for symbolic matrix computation in Python. We show that LSMMD-MA scales to a million cells in each modality, two orders of magnitude greater than existing implementations. LSMMD-MA is freely available at https://github.com/google-research/large_scale_mmdma and archived at https://doi.org/10.5281/zenodo.8076311.

Image Encryption and Decryption Using Vigenere Cipher with Compute Unified Device Architecture (CUDA)

Compute Unified Device Architecture (CUDA) adalah Application Programming Interface (API) NVIDIA dan platform yang memungkinkan akses langsung ke set instruksi GPU dan memberi dukungan untuk berinteraksi dengan GPU terkait komputasi paralel. Dengan CUDA, komputasi yang kompleks menjadi lebih cepat dan lebih efisien. Vigenere Cipher adalah kriptografi klasik populer yang mengimplementasikan kunci simetris dengan panjang tertentu. Pada penelitian ini, penerapan enkripsi dan dekripsi Vigenere Cipher dilakukan pada citra serta dengan CPU dan GPU (CUDA). Paralelisasi dengan CUDA menunjukkan hasil eksekusi waktu yang relatif lebih cepat daripada CPU. Persentase rata-rata penurunan waktu adalah 99,46 persen untuk enkripsi serta 99,47 persen untuk dekripsi.Persentase rata-rata penurunan waktu adalah 99,46 persen untuk enkripsi serta 99,47persen untuk dekripsi.

A parallel particle swarm optimization algorithm based on GPU/CUDA

Parallel computing is the main way to improve the computational efficiency of metaheuristic algorithms for solving high-dimensional, nonlinear optimization problems. Previous studies have typically only implemented local parallelism for the particle swarm optimization (PSO) algorithm. In this study, we proposed a new parallel particle swarm optimization algorithm (GPU-PSO) based on the Graphics Processing Units (GPU) and Compute Unified Device Architecture (CUDA), which uses a combination of coarse-grained parallelism and fine-grained parallelism to achieve global parallelism. In addition, we designed a data structure based on CUDA features and utilized a merged memory access mode to further improve data-parallel processing and data access efficiency. Experimental results show that the algorithm effectively reduces the solution time of PSO for solving high-dimensional, large-scale optimization problems. The speedup ratio increases with the dimensionality of the objective function, where the speedup ratio is up to 2000 times for the high-dimensional Ackley function.

Performance Investigation of the Conjunction Filter Methods and Enhancement of Computation Speed on Conjunction Assessment Analysis with CUDA Techniques

The growing number of space objects leads to increases in the potential risks of damage to satellites and generates space debris after colliding. Conjunction assessment analysis is the one of keys to evaluating the collision risk of satellites and satellite operators require the analyzed results as fast as possible to decide and execute collision maneuver planning. However, the computation time to analyze the potential risk of all satellites is proportional to the number of space objects. The conjunction filters and parallel computing techniques can shorten the computation cost of conjunction analysis to provide the analyzed results. Therefore, this paper shows the investigation of the conjunction filter performances (accuracy and computation speed): Smart Sieve, CSieve and CAOS-D (combination of both Smart Sieve and CSieve) in both a single satellite (one vs. all) and all space objects (all vs. all) cases. Then, all the screening filters are developed to implement an algorithm that executes General-purpose computing on graphics processing units (GPGPU) by using NVIDIAs Compute Unified Device Architecture (CUDA). The analyzed results show the comparison results of the accuracy of conjunction screening analysis and computation times of each filter when implemented with the parallel computation techniques.

MCR toolkit: A GPU-based toolkit for multi-channel reconstruction of preclinical and clinical x-ray CT data.

The advancement of x-ray CT into the domains of photon counting spectral imaging and dynamic cardiac and perfusion imaging has created many new challenges and opportunities for clinicians and researchers. To address challenges such as dose constraints and scanning times while capitalizing on opportunities such as multi-contrast imaging and low-dose coronary angiography, these multi-channel imaging applications require a new generation of CT reconstruction tools. These new tools should exploit the relationships between imaging channels during reconstruction to set new image quality standards while serving as a platform for direct translation between the preclinical and clinical domains. We outline and demonstrate a new Multi-Channel Reconstruction (MCR) Toolkit for GPU-based analytical and iterative reconstruction of preclinical and clinical multi-energy and dynamic x-ray CT data. To promote open science, open-source distribution of the Toolkit will coincide with the release of this publication (GPL v3; gitlab.oit.duke.edu/dpc18/mcr-toolkit-public). The MCR Toolkit source code is implemented in C/C++ and NVIDIA's CUDA GPU programming interface, with scripting support from MATLAB and Python. The Toolkit implements matched, separable footprint CT reconstruction operators for projection and backprojection in two geometries: planar, cone-beam CT (CBCT) and 3rd generation, cylindrical multi-detector row CT (MDCT). Analytical reconstruction is performed using filtered backprojection (FBP) for circular CBCT, weighted FBP (WFBP) for helical CBCT, and cone-parallel projection rebinning followed by WFBP for MDCT. Arbitrary combinations of energy and temporal channels are iteratively reconstructed under a generalized multi-channel signal model for joint reconstruction. We solve this generalized model algebraically using the split Bregman optimization method and the BiCGSTAB(l) linear solver interchangeably for both CBCT and MDCT data. Rank-sparse kernel regression (RSKR) and patch-based singular value thresholding (pSVT) are used to regularize the energy and time dimensions, respectively. Under a Gaussian noise model, regularization parameters are estimated automatically from the input data, dramatically reducing algorithm complexity for end users. Multi-GPU parallelization of the reconstruction operators is supported to manage reconstruction times. Denoising with RSKR and pSVT and post-reconstruction material decomposition are illustrated with preclinical and clinical cardiac photon-counting (PC)CT data. A digital MOBY mouse phantom with cardiac motion is used to illustrate single energy (SE), multi-energy (ME), time resolved (TR), and combined multi-energy and time-resolved (METR) helical, CBCT reconstruction. A fixed set of projection data is used across all reconstruction cases to demonstrate the Toolkit's robustness to increasing data dimensionality. Identical reconstruction code is applied to in vivo cardiac PCCT data acquired in a mouse model of atherosclerosis (METR). Clinical cardiac CT reconstruction is illustrated using the XCAT phantom and the DukeSim CT simulator, while dual-source, dual-energy CT reconstruction is illustrated for data acquired with a Siemens Flash scanner. Benchmarking results with NVIDIA RTX 8000 GPU hardware demonstrate 61%-99% efficiency in scaling computation from one to four GPUs for these reconstruction problems. The MCR Toolkit provides a robust solution for temporal and spectral x-ray CT reconstruction problems and was built from the ground up to facilitate translation of CT research and development between preclinical and clinical applications.

CuCatch: A Debugging Tool for Efficiently Catching Memory Safety Violations in CUDA Applications

CUDA, OpenCL, and OpenACC are the primary means of writing general-purpose software for NVIDIA GPUs, all of which are subject to the same well-documented memory safety vulnerabilities currently plaguing software written in C and C++. One can argue that the GPU execution environment makes software development more error prone. Unlike C and C++, CUDA features multiple, distinct memory spaces to map to the GPU’s unique memory hierarchy, and a typical CUDA program has thousands of concurrently executing threads. Furthermore, the CUDA platform has fewer guardrails than CPU platforms that have been forced to incrementally adjust to a barrage of security attacks. Unfortunately, the peculiarities of the GPU make it difficult to directly port memory safety solutions from the CPU space. This paper presents cuCatch, a new memory safety error detection tool designed specifically for the CUDA programming model. cuCatch combines optimized compiler instrumentation with driver support to implement a novel algorithm for catching spatial and temporal memory safety errors with low performance overheads. Our experimental results on a wide set of GPU applications show that cuCatch incurs a 19% runtime slowdown on average, which is orders of magnitude faster than state-of-the-art debugging tools on GPUs. Moreover, our quantitative evaluation demonstrates cuCatch’s higher error detection coverage compared to prior memory safety tools. The combination of high error detection coverage and low runtime overheads makes cuCatch an ideal candidate for accelerating memory safety debugging for GPU applications.

Research on Seismic Connectivity Reliability Analysis of Water Distribution System Based on CUDA

To improve the seismic connectivity reliability (SCR) analysis efficiency of water distribution systems (WDS) based on Monte Carlo (MC) simulation, the quasi-Monte Carlo (QMC) method sampled by a low-discrepancy sequence is applied. Furthermore, a parallel algorithm combined with the breadth-first search algorithm for SCR analysis of WDS based on the QMC method and Compute Unified Device Architecture (CUDA) platform was proposed. A city WDS was taken as a computational example, the accuracy and efficiency of the traditional MC algorithm and parallel algorithm were compared, and the influence of the Sobol sequence and pseudo-random number sequence was analysed. The analysis results show that when 1,000,000 simulations are performed, the maximum error of the calculation results of the two methods is 0.2%, and the parallel method can obtain a six-fold speedup ratio compared with the serial method, indicating that the proposed parallel method is correct, meets the accuracy requirements, and helps to improve the SCR analysis efficiency. When the number of simulations is the same, the simulation results based on the Sobol sequence are more accurate than those based on the pseudo-random number sequence. The proposed parallel method also achieves a good acceleration effect in the SCR analysis of large-scale WDS.

Adaptation of FPGA architecture for accelerated image preprocessing

The work is devoted to the topical problem at the intersection of communications theory, digital electronics and numerical analysis, namely the study of image processing methods implementation time on different architectures of computational devices, which are used for software and hardware acceleration. The subject of this article is the investigation of reconfigurable FPGA processing systems in the image processing area. The goal of this work is to create a reconfigurable FPGA-based image processing system and compare it with existing processing architectures. Task. To fulfill the requirements of this work, it is necessary to prepare a practical experiment as well as theoretical research of the proposed architecture; to investigate the process of creating a ZYNQ SoC-based image processing system; and to develop and benchmark the speed of execution for the given set of algorithms with the specific range of the picture resolution. Methods used: FPGA simulation, C++ parallel programming with OpenMP, NVIDIA CUDA, performance analysis tools. The result of this work is the development of a resilient SoC Zynq7000–based computing system with programmable logic and the possibility to load images to FPGA RAM using the resources of ARM core for further processing and output via HDMI video interface, which enables the change of PL configuration at any time during the processing process. Conclusions. The efficiency of the FPGA approach was compared with a parallel image processing method implementation with OpenMP and CUDA. An overview of the ZYNQ platform with specific details related to media processing is presented. The analysis of algorithm speed testing findings based on various outputs proved the advantage (of over 60 times) of hardware acceleration of image processing over software analogs. The obtained results may be used in the development of embedded SoC-based solutions that require acceleration of big data processing. Also, the achieved findings can be used during the process of finding a suitable embedded platform for a certain image-processing task, where high data throughput is one of the most desired requirements.

Colour-weighted rank transform and improved dynamic programming for fast and accurate stereo matching

ABSTRACT Real-time stereo matching with high accuracy is a dynamic research topic; it is attractive in diverse computer vision applications. This paper presents a stereo-matching algorithm that produces high-quality disparity map while maintaining real-time performance. The proposed stereo-matching method is based on three per-pixel difference measurements with adjustment elements. The absolute differences and the gradient matching are combined with a colour-weighted extension of complete rank transform to reduce the effect of radiometric distortion. The disparity calculation is realized using improved dynamic programming that optimizes along and across all scanlines. It solves the inter-scanline inconsistency problem and increases the matching accuracy. The proposed algorithm is implemented on parallel high-performance graphic hardware using the Compute Unified Device Architecture to reach over 240 million disparity evaluations per second. The processing speed of our algorithm reaches 98 frames per second on 240 × 320-pixel images and 32 disparity levels. Our method ranks fourth in terms of accuracy and runtime for quarter-resolution images in the Middlebury stereo benchmark.

ESTIMATION OF POWER CONSUMPTION OF MOBILE DEVICES IN CLOUD COMPUTING

Modern computing tasks require an increase in computing power. This necessitates the creation and production of new equipment for cloud computing. At the same time, the number of personal mobile devices is already measured in billions, and even their partial use could reduce production requirements. In addition, mobile hardware is more energy efficient, which contributes to significant energy savings. The article investigates the issue of qualitative and quantitative assessment of the efficiency of using mobile devices for computing compared to traditional stationary solutions. The purpose of the work is to substantiate the following hypothesis: computing in the cloud based on mobile devices significantly reduces energy consumption than computing on stationary equipment. For this purpose, we show that computing on a mobile GPU is more energy efficient than computing on a stationary processor. Public sources and benchmarks were analyzed to determine the qualitative advantage. On the basis of the studied data, efficiency indicators for various mobile and desktop GPUs are calculated. It is argued that in most cases, mobile solutions consume significantly less energy compared to desktop solutions. To calculate the quantitative advantage, an experiment was conducted on the basis of two platforms: mobile and desktop. The same computational task was implemented using Apple Metal and NVidia CUDA. Based on this task, the energy efficiency indicators of the mobile and stationary graphic professor were calculated. According to the results of the study, a significant advantage of the mobile GPU in terms of energy efficiency has been determined. This result is relevant because the platforms were released in the same year with a difference of several months, so they can be considered peers of each other. The approaches presented here do not take into account the consumption of all other parts of the system, except for the GPUs. This means that the consumption of the motherboard, power supply, etc. can tilt the balance in favor of the mobile processor even more. But for distributed computing, the network connection is very important, and it can consume a significant amount of power on a mobile device. Further research will focus on a more comprehensive accounting of the energy consumption of various computer subsystems.

Biometric face recognition method using graphics processing unit system

The expansion of biometric applications and databases is worrying. Processing extensive or sophisticated biometric data results in longer wait times, which might restrict application usefulness. This work focuses on accelerating the processing of biometric data and proposes a parallel method of data processing that exceeds the capabilities of a central processing unit (CPU). The combination of the graphics processing unit (GPU) and compute unified device architecture (CUDA) results in at least three times the processing speed of a published accurate and secure multimodal biometric system. The GPU-assisted approach beats the CPU-only implementation when saturating the CPU-only performance with more people than the available thread count. The GPU-assisted solution is also proven to have the same accuracy as the original system, indicating accuracy and processing performance improvements in the demanding big data environment.

A Numerical Study of an Ellipsoidal Nanoparticles under High Vacuum Using the DSMC Method

The semiconductor and display manufacturing process requires high precision. Therefore, inside the equipment, fine impurity particles affect the yield rate of production. However, since most manufacturing processes are performed under high-vacuum conditions, it is difficult to estimate particle flow with conventional analytical tools. In this study, high-vacuum flow was analyzed using the direct simulation Monte Carlo (DSMC) method, and various forces acting on fine particles in a high-vacuum flow field were calculated. To compute the computationally intensive DSMC method, GPU-based computer unified device architecture (CUDA) technology was used. The force acting on the particles in the high-vacuum rarefied gas region was verified using the results of previous studies, and the results were derived for the difficult-to-experiment region. An ellipsoid shape with an aspect ratio rather than a spherical shape was also analyzed. The change in drag force according to various aspect ratios was analyzed and compared with the results of the spherical shape under the same flow conditions.

The VM2D Open Source Code for Two-Dimensional Incompressible Flow Simulation by Using Fully Lagrangian Vortex Particle Methods

This article describes the open-source C++ code VM2D for the simulation of two-dimensional viscous incompressible flows and solving fluid-structure interaction problems. The code is based on the Viscous Vortex Domains (VVD) method developed by Prof. G. Ya. Dynnikova, where the viscosity influence is taken into account by introducing the diffusive velocity. The original VVD method was supplemented by the author’s algorithms for boundary condition satisfaction, which made it possible to increase the accuracy of flow simulation near the airfoil’s surface line and reduce oscillations when calculating hydrodynamic loads. This paper is aimed primarily at assessing the efficiency of the parallelization of the algorithm. OpenMP, MPI, and Nvidia CUDA parallel programming technologies are used in VM2D, which allow performing simulations on computer systems of various architectures, including those equipped with graphics accelerators. Since the VVD method belongs to the particle methods, the efficiency of parallelization with the usage of graphics accelerators turns out to be quite high. It is shown that in a real simulation, one graphics card can replace about 80 nodes, each of which is equipped with 28 CPU cores. The source code of VM2D is available on GitHub under GNU GPL license.

Three-dimensional third-order gas-kinetic scheme on hybrid unstructured meshes for Euler and Navier–Stokes equations

In this paper, a third-order gas-kinetic scheme is developed on the three-dimensional hybrid unstructured meshes for the compressible inviscid and viscous flows. In the classical weighted essentially non-oscillatory (WENO) scheme, the high-order spatial accuracy is achieved by the non-linear combination of lower order polynomials. However, for the hybrid unstructured meshes, the procedures, including the selection of candidate stencils and calculation of linear weights, become extremely complicated, especially for three-dimensional problems. To overcome the drawbacks, a third-order WENO reconstruction is developed on the hybrid unstructured meshes, including tetrahedron, pyramid, prism and hexahedron. A general strategy is given for the selection of big stencil and sub-stencils for hybrid meshes, and the topologically independent linear weights are used in the spatial reconstruction. A unified interpolation is also used for the volume integral of different types of control volumes, as well as the flux integration over different cell interfaces. Incorporate with the two-stage fourth-order temporal discretization, the explicit high-order gas-kinetic schemes are developed for unsteady problems. With the lower–upper symmetric Gauss–Seidel (LU-SGS) methods and Jacobi iteration, the implicit high-order gas-kinetic schemes are developed for steady problems. To accelerate the computation, both explicit and implicit scheme is implemented with the graphics processing unit (GPU) using compute unified device architecture (CUDA). Especially, for implicit parallel computation, Jacobi iteration is used. The speedup of GPU code suggests the potential of high-order gas-kinetic schemes for the large scale computation. A variety of numerical examples, from the subsonic to supersonic flows, are presented to validate the accuracy and robustness for both inviscid and viscous flows.

Accelerated Peridynamic Computation on GPU for Quasi-static Fracture Simulations

Due to the adoption of integral equations instead of partial differential equations to describe the deformations of materials, the peridynamics has advantages in dealing with fracture problems. However, the non-local effect of peridynamics brings too much computation cost to make large-scale engineering applications challenging. In this paper, a parallel algorithm in the Compute Unified Device Architecture (CUDA) framework is presented to speed up the computational process of a peridynamic model for quasistatic fracture simulations on GPU, in which the peridynamic model is numerically implemented by the peridynamics-based finite element method (PeriFEM) [1]. The parallel algorithm makes crack simulations by the pure peridynamics with millions of degrees of freedom possible with one GPU. To validate the accuracy and efficiency of the parallel algorithm based on PeriFEM, several numerical benchmarks are performed, and the results are compared with those obtained by the finite element method (FEM) and the serial algorithm. The results of the comparison show that the presented parallel algorithm is effective and efficient.

Spinsim: A GPU optimized python package for simulating spin-half and spin-one quantum systems

Spinsim simulates the quantum dynamics of unentangled spin-1/2 and spin-1 systems evolving under time-dependent control. While other solvers for the time-dependent Schrödinger equation optimize for larger state spaces but less temporally-rich control, spinsim is optimized for intricate time evolution of a minimalist system. Efficient simulation of individual or ensemble quanta driven by adiabatic sweeps, elaborate pulse sequences, complex signals and non-Gaussian noise is the primary target application. We achieve fast and robust evolution using a geometric integrator to bound errors over many steps, and split the calculation parallel-in-time on a GPU using the numba just-in-time compiler. Speed-up is three orders of magnitude over QuTip's sesolve and Mathematica's NDSolve, and four orders over SciPy's ivp_solve for equal accuracy. Interfaced through python, spinsim should be useful for simulating robust state preparation, inversion and dynamical decoupling sequences in NMR and MRI, and in quantum control, memory and sensing applications with two- and three-level quanta. Program summaryProgram Title: SpinsimCPC Library link to program files:https://doi.org/10.17632/f6rdk4gyxr.1Developer's repository link:https://github.com/alexander-tritt-monash/spinsimLicensing provisions: BSD 3-clauseProgramming language: Python (3.7 or greater)Nature of problem: Quantum sensing is a domain of quantum technology where the dynamics of quantum systems are used to infer properties of the systems' environments. The development of quantum sensing protocols is greatly sped-up by software simulations of the new techniques. Quantum sensing simulation benefits from temporally-rich control of individual quanta. However, current specialized time-dependent Schrödinger equation solvers are instead optimized only for simple pulses in large Hilbert spaces. Thus, there is a need for efficient simulation of individual or ensemble quanta driven by adiabatic sweeps, elaborate pulse sequences, complex signals and non-Gaussian noise.Solution method: Spinsim simulates the quantum dynamics of spin-1/2 and spin-1 systems evolving under time-dependent control. We first speed up the integration of the time-dependent Schrödinger equation by splitting the calculation parallel-in-time on a GPU using the numba [1] just-in-time compiler. We achieve fast and robust evolution using a geometric integrator to bound errors over many steps. A dynamic rotating frame transformation and Lie-Trotter decomposition are used to decrease effective step sizes on long and short time scales, respectively. Hence, each individual step is more accurate. Spinsim is interfaced via a python package, meaning it can be used by researchers inexperienced with geometric integrators and GPU parallelism.Additional comments including restrictions and unusual features: To use the (default) Nvidia cuda GPU parallelization, one needs to have a cuda compatible Nvidia GPU [2]. For cuda mode to function, one also needs to install the Nvidia cuda toolkit [3]. If cuda is not available on the system, the simulator will automatically parallelize over multicore CPUs instead. Developed and tested on Windows 10; tested on Windows 11. CPU functionality tested on MacOS 10.16 Big Sur (note that MacOS 10.14 Mojave and higher are not compatible with cuda hardware and software). The package (including cuda functionality) is in principle compatible with Linux, but functionality has not been tested.

APPROACH FOR CODE ANALYSIS TO ESTIMATE POWER CONSUMPTION OF CUDA CORE

The graphics processing unit is a popular computing device for achieving exascale performance in high-performance computing programs, which is used not only in graphics tasks, but also in computational tasks such as machine learning, scientific computing, and cryptography. With the help of a graphics processor, you can achieve significant speed and performance compared to the central processing unit. CUDA, Compute Unified Device Architecture, a graphics processing unit software development platform, allows developers to use the high-performance computing capabilities of graphics processing units to solve problems traditionally handled by central processing units. Even though the graphics processing unit has a relatively high power to performance ratio, it consumes a significant amount of power during computing. The paper proposes an approach for code analysis to estimate power consumption of CUDA core to improve the power efficiency of applications focused on computing on graphics processing units. The proposed approach makes it possible to estimate the power consumption of such applications without the need to run them on physical devices. The proposed approach is based on static analysis of the CUDA program and machine learning methods. To evaluate the effectiveness of the proposed approach, three graphics processing unit architectures were used: NVIDIA PASCAL, NVIDIA TURING, and NVIDIA AMPERE. The results of the experiments showed that for the NVIDIA AMPERE architecture, the proposed approach using decision trees makes it possible to achieve a determination coefficient of 0.9173. The results obtained confirm the effectiveness of the proposed code analysis method for estimating the power consumption of the CUDA core. This method can be useful for CUDA developers who want to improve the efficiency and power efficiency of their programs.

Enable high-resolution, real-time ensemble simulation and data assimilation of flood inundation using distributed GPU parallelization

Numerical modeling of the intensity and evolution of flood events are affected by multiple sources of uncertainty such as precipitation and land surface conditions. To quantify and curb these uncertainties, an ensemble-based simulation and data assimilation model for pluvial flood inundation is constructed. The shallow water equation is decoupled in the x and y directions, and the inertial form of the Saint-Venant equation is chosen to realize fast computation. The probability distribution of the input and output factors is described using Monte Carlo samples. Subsequently, a particle filter is incorporated to enable the assimilation of hydrological observations and improve prediction accuracy. To achieve high-resolution, real-time ensemble simulation, heterogeneous computing technologies based on CUDA (compute unified device architecture) and a distributed storage multi-GPU (graphics processing unit) system are used. Multiple optimization skills are employed to ensure the parallel efficiency and scalability of the simulation program. Taking an urban area of Fuzhou, China as an example, a model with a 3-m spatial resolution and 4.0 million units is constructed, and 8 Tesla P100 GPUs are used for the parallel calculation of 96 model instances. Under these settings, the ensemble simulation of a 1-hour hydraulic process takes 2.0 min, which achieves a 2680 × estimated speedup compared with a single-thread run on CPU. The calculation results indicate that the particle filter method effectively constrains simulation uncertainty while providing the confidence intervals of key hydrological elements such as streamflow, submerged area, and submerged water depth. The presented approaches show promising capabilities in handling the uncertainties in flood modeling as well as enhancing prediction efficiency.