Hybrid CPU Research Articles

Direct reduction of iron-ore with hydrogen in fluidized beds: A coarse-grained CFD-DEM-IBM study

Hydrogen metallurgy technology uses hydrogen as the reducing agent instead of carbon reduction, which is one of the important ways to reduce carbon dioxide emissions and ensure the green and sustainable development of iron and steel industry. Due to the advantages of high gas-solid contact efficiency and outstanding mass and heat transfer, direct reduction of iron ore in fluidized beds has attracted much attention. Based on the three-layer unreacted shrinking core model and considering the diversity of particle structure during the reaction process, a multi-reaction path (multi-resistance network) model was established. In this study, a coarse-grained CFD-DEM-IBM solver based on hybrid CPU–GPU computing is developed to simulate the direct reduction process of two kinds of iron ore with hydrogen in fluidized beds, where the developed unreacted shrinking core model is used to model the reduction reactions, a coarse-grained model and multiple GPUs enable the significant acceleration of particle computation, and the immersed boundary method (IBM) enables the use of simple mesh even in complex geometries of reactors. The predicted results of particle reduction degree are in good agreement with the experimental values, which proves the correctness of the CFD-DEM-IBM solver. In addition, the effects of reaction kinetic parameters and operating temperature on particle reduction degree are also investigated. Present study provides a method for digital design, optimization and scale-up of ironmaking reactors.

A comparison of Algebraic Multigrid Bidomain solvers on hybrid CPU–GPU architectures

The numerical simulation of cardiac electrophysiology is a highly challenging problem in scientific computing. The Bidomain system is the most complete mathematical model of cardiac bioelectrical activity. It consists of an elliptic and a parabolic partial differential equation (PDE), of reaction–diffusion type, describing the spread of electrical excitation in the cardiac tissue. The two PDEs are coupled with a stiff system of ordinary differential equations (ODEs), representing ionic currents through the cardiac membrane. Developing efficient and scalable preconditioners for the linear systems arising from the discretization of such computationally challenging model is crucial in order to reduce the computational costs required by the numerical simulations of cardiac electrophysiology. In this work, focusing on the Bidomain system as a model problem, we have benchmarked two popular implementations of the Algebraic Multigrid (AMG) preconditioner embedded in the PETSc library and we have studied the performance on the calibration of specific parameters. We have conducted our analysis on modern HPC architectures, performing scalability tests on multi-core and multi-GPUs settings. The results have shown that, for our problem, although scalability is verified on CPUs, GPUs are the optimal choice, since they yield the best performance in terms of solution time.

On floating point precision in computational fluid dynamics using OpenFOAM

Thanks to the computational power of modern cluster machines, numerical simulations can provide, with an unprecedented level of details, new insights into fluid mechanics. However, taking full advantage of this hardware remains challenging since data communication remains a significant bottleneck to reaching peak performances. Reducing floating point precision is a simple and effective way to reduce data movement and improve the computational speed of most applications. Nevertheless, special care needs to be taken to ensure the quality and convergence of computed solutions, especially when dealing with complex fluid simulations. In this work, we analyse the impact of reduced (single and mixed compared to double) precision on computational performance and accuracy for computational fluid dynamics. Using the open source library OpenFOAM, we consider incompressible, compressible, and multiphase fluid solvers for testing on relevant benchmarks for flows in the laminar and turbulent regime and in the presence of shock waves. Computational gain and changes in the scalability of applications in reduced precision are also discussed. In particular, an ad hoc theoretical model for the strong scaling allows us to interpret and understand the observed behaviours, as a function of floating point precision and hardware specifics. Finally, we show how reduced precision can significantly speed up a hybrid CPU–GPU implementation, made available to OpenFOAM end-users recently, that simply relies on a GPU linear algebra solver developed by hardware vendors.

Quantum AI simulator using a hybrid CPU–FPGA approach

The quantum kernel method has attracted considerable attention in the field of quantum machine learning. However, exploring the applicability of quantum kernels in more realistic settings has been hindered by the number of physical qubits current noisy quantum computers have, thereby limiting the number of features encoded for quantum kernels. Hence, there is a need for an efficient, application-specific simulator for quantum computing by using classical technology. Here we focus on quantum kernels empirically designed for image classification and demonstrate a field programmable gate arrays (FPGA) implementation. We show that the quantum kernel estimation by our heterogeneous CPU–FPGA computing is 470 times faster than that by a conventional CPU implementation. The co-design of our application-specific quantum kernel and its efficient FPGA implementation enabled us to perform one of the largest numerical simulations of a gate-based quantum kernel in terms of features, up to 780-dimensional features. We apply our quantum kernel to classification tasks using the Fashion-MNIST dataset and show that our quantum kernel is comparable to Gaussian kernels with the optimized hyperparameter.

A hybrid CPU–GPU solver based on three-dimensional spectral Chebyshev technique for determining the dynamic behavior of thick sandwich panels

In this study, static and vibration behavior of a curved thick sandwich-structured composite composed of a honeycomb core layer and two face-sheets reinforced with carbon nanotubes is investigated. The governing equations are derived using three-dimensional elasticity equations and Hamilton’s principle. The problem domain is discretized following Gauss–Lobatto sampling approach. To numerically solve the governing equations, a spectral method based on Chebyshev polynomials is used and a CPU–GPU based hybrid solver is developed. The presented approach is validated by comparing the predicted natural frequencies and deformation of the structure to those obtained from finite element analysis. The results show that in-house CPU–GPU based spectral solution approach decreases the computational duration remarkably and predicts the dynamic and static behavior of the system accurately. Since composite materials offer great flexibility in tailoring the dynamic behavior of the structure, an optimization study is also performed to benefit from the developed computationally efficient solver. In this study, the CNT orientations are selected as design variables to maximize the fundamental frequency of the structure. Based on the analyzed cases for different thickness ratios, curvature amounts, volume content, and distributions of CNTs, it is shown that the vibration behavior of the structure can be significantly tailored.

A Real-Time Ultra-High Definition Video Decoder of AVS3 on Heterogeneous Systems

Recent years have witnessed the exponential increase in the demand for ultra high definition (UHD) video compression. The third generation of Audio Video Coding Standard (AVS3), which is also known as IEEE Standard 1857.10, is the latest audio and video coding standard developed by the China AVS working group. In AVS3, targeting for UHD videos, a series of efficient coding tools have been introduced, leading to the dramatical increase of computational burden. In this scenario, real-time decoding of UHD videos becomes extremely challenging. This paper presents an improved hybrid CPU + GPU accelerated framework for AVS3 decoding. In particular, the motion vector (MV) derivation process is extracted from entropy decoding threads on the CPU. Therefore, the dependency between threads is removed and the entropy decoding can be performed by multiple threads efficiently. Regarding the GPU, we design compact data structures for transform, prediction, and in-loop filtering to reduce the burden of data transmission. A flexible information buffer supporting multi-thread random writing is further created to coordinate the computation between CPU and GPU. Through asynchronous operations on the buffer, the computation of different computing units and the data transmission between them could be performed in parallel. With NVIDIA GeForce RTX 2080Ti GPU and Intel Core i7 8700K CPU, the proposed decoder achieves 151 frames per second (fps) for 4K videos and 55 fps for 8K videos in all intra configuration. In random access configuration, 218 fps and 74 fps are obtained for 4K and 8K videos, respectively.

Hybrid CPU Scheduling algorithm SJF-RR In Static Set of Processes

Round Robin (RR) algorithm is widely used in modern operating systems (OS) as it has a better responsiveness as periodic quantum (occurring at regular intervals) in addition to have a good feature such as low scheduling overhead of n processes in a ready queue which takes a constant time O(1). But, RR algorithms have the worse features such as having low throughput, long average turnaround and waiting time, in addition to the number of context switches for (n) processes is (n) switches. Shortest Job First (SJF) however, itself is not practical in time sharing Oss due to its low response. More over the scheduling overhead of n processes in a ready queue is O(n), But the good features of SJF algorithm are the best average turnaround time and waiting time.By considering a static set of n processes, desirable features of CPU scheduler to maximize CPU utilization, response time and minimize waiting time and turnaround time are obtained by combining the kernel properties of SJF algorithm with the best features of RR algorithm to produce a new algorithm as an original and novel algorithm called; " Hybrid CPU Scheduling algorithm SJF-RR in Static Set of Processes " which, proposed in this research.The proposed algorithm is implemented through an innovative optimal equation to adapt time quantum factor for each process in each round as a periodic quantum (occurred at irregular intervals). That is while applying proposed algorithm, mathematical calculations take place to have particular time quantum for each process. Once, a criterion has been selected for comparison, deterministic modeling with the same numbers for input is proven that proposed algorithm is the best.

Enabling simulation of high‐dimensional micro‐macro biophysical models through hybrid CPU and multi‐GPU parallelism

AbstractMicro‐macro models provide a powerful tool to study the relationship between microscale mechanisms and emergent macroscopic behavior. However, the detailed microscopic modeling requires tracking and evolving a potentially high‐dimensional configuration space at high computational cost. In this work, we present a novel parallel algorithm for simulating a high‐dimensional micro‐macro model of a gliding motility assay. We utilize a holistic approach aligning the data residency and simulation scales with the hybrid CPU and multi‐GPU hardware. Our novel approach achieves a speedup factor of 9.25× over previous GPU‐accelerated micro‐macro methods on the same hardware. Furthermore, by decoupling dependencies in the microstructure update, we are able to efficiently distribute the microstructure over multiple GPUs with minimal overhead. We test on up to four GPUs and observe excellent scaling, suggesting that significant further speedups are achievable with additional GPUs. Our approach enables micro‐macro simulations of higher complexity and resolution than would otherwise be feasible.

GPU-acceleration of the ELPA2 distributed eigensolver for dense symmetric and hermitian eigenproblems

The solution of eigenproblems is often a key computational bottleneck that limits the tractable system size of numerical algorithms, among them electronic structure theory in chemistry and in condensed matter physics. Large eigenproblems can easily exceed the capacity of a single compute node, thus must be solved on distributed-memory parallel computers. We here present GPU-oriented optimizations of the ELPA two-stage tridiagonalization eigensolver (ELPA2). On top of cuBLAS-based GPU offloading, we add a CUDA kernel to speed up the back-transformation of eigenvectors, which can be the computationally most expensive part of the two-stage tridiagonalization algorithm. We benchmark the performance of this GPU-accelerated eigensolver on two hybrid CPU–GPU architectures, namely a compute cluster based on Intel Xeon Gold CPUs and NVIDIA Volta GPUs, and the Summit supercomputer based on IBM POWER9 CPUs and NVIDIA Volta GPUs. Consistent with previous benchmarks on CPU-only architectures, the GPU-accelerated two-stage solver exhibits a parallel performance superior to the one-stage counterpart. Finally, we demonstrate the performance of the GPU-accelerated eigensolver developed in this work for routine semi-local KS-DFT calculations comprising thousands of atoms.

Digital blood in massively parallel CPU/GPU systems for the study of platelet transport.

We propose a highly versatile computational framework for the simulation of cellular blood flow focusing on extreme performance without compromising accuracy or complexity. The tool couples the lattice Boltzmann solver Palabos for the simulation of blood plasma, a novel finite-element method (FEM) solver for the resolution of deformable blood cells, and an immersed boundary method for the coupling of the two phases. The design of the tool supports hybrid CPU–GPU executions (fluid, fluid–solid interaction on CPUs, deformable bodies on GPUs), and is non-intrusive, as each of the three components can be replaced in a modular way. The FEM-based kernel for solid dynamics outperforms other FEM solvers and its performance is comparable to state-of-the-art mass–spring systems. We perform an exhaustive performance analysis on Piz Daint at the Swiss National Supercomputing Centre and provide case studies focused on platelet transport, implicitly validating the accuracy of our tool. The tests show that this versatile framework combines unprecedented accuracy with massive performance, rendering it suitable for upcoming exascale architectures.

Online multimedia retrieval on CPU–GPU platforms with adaptive work partition

Nearest neighbors search is a core operation found in several online multimedia services. These services have to handle very large databases, while, at the same time, they must minimize the query response times observed by users. This is specially complex because those services deal with fluctuating query workloads (rates). Consequently, they must adapt at run-time to minimize the response times as the load varies. In this paper, we address the aforementioned challenges with a distributed memory parallelization of the product quantization nearest neighbor search, also known as IVFADC, for hybrid CPU–GPU machines. Our parallel IVFADC implements an out-of-GPU memory execution scheme to use the GPU for databases in which the index does not fit in its memory, which is crucial for searching in very large databases. The careful use of CPU and GPU with work stealing led to an average response time reduction of 2.4× as compared to using the GPU only. Also, our approach to adapt the system to fluctuating loads, called Dynamic Query Processing Policy (DQPP), attained a response time reduction of up to 5× vs. the best static (BS) policy for moderate loads. The system has attained high query processing rates and near-linear scalability in all experiments. We have evaluated our system on a machine with up to 256 NVIDIA V100 GPUs processing a database of 256 billion SIFT features vectors.

Runtime power allocation approach for GAMESS hybrid CPU‐GPU implementation

SummaryTo improve power consumption of applications at the runtime, modern processors provide frequency scaling capabilities, which along with workload optimization, are also available on GPU accelerators. In this work, a runtime strategy is proposed to distribute a given power allocation among the host components and the GPU according to the current application performance and power usage, such that GPU execution is prioritized over CPU for power allocation to maximize application performance. Next, the strategy is tailored to an application, a quantum‐chemistry package GAMESS for ab initio electronic structure calculations. Specifically, GAMESS hybrid CPU–GPU implementation as provided in the Libcchem library is considered. Experiments, performed on a 28‐core node with a Kepler GPU, resulted in performance gains of up to 50% under the proposed strategy and the largest power allocation considered here as compared with the scenario when this allocation was equally distributed among the computing‐platform components.

Knowledge based fuzzy c-means method for rapid brain tissues segmentation of magnetic resonance imaging scans with CUDA enabled GPU machine

Fuzzy C-Means (FCM) plays a major role in brain tissue segmentation. The proposed method aims to implements rapid brain tissue segmentation from MRI human head scans using FCM in CPU and GPU. This method is known as FCM-GENIUS. This paper presents three novel steps to enrich the performance of conventional FCM algorithm in CPU. There are region of interest (ROI) selection, knowledge based initialization and knowledge based optimization. The ROI selection is a preprocessing step contains brain extraction and bounding box processes. An automatic knowledge based initialization to FCM algorithm using histogram smoothing for centroids selection from middle slice of the given MRI brain volume. Optimization helps to improve the computation speed up of FCM algorithm using MRI slice adjacency property. The materials used for the proposed work are gathering from internet brain segmentation repository (IBSR). The accuracy of segmentation also compared with traditional and existing methods. The proposed method yield equal segmentation accuracy compared with existing methods but reduces the segmentation time considerably up to seven times and average number of iterations up to three times. In addition, parallel FCM implements in GPU machine and the performance was compared with the conventional FCM in CPU. The single instruction multiple data (SIMD) model was used with the hybrid CPU–GPU implementation in the GPU machine to accelerate the medical image segmentation.

Cosmic Dawn II (CoDa II): a new radiation-hydrodynamics simulation of the self-consistent coupling of galaxy formation and reionization

ABSTRACT Cosmic Dawn II (CoDa II) is a new, fully coupled radiation-hydrodynamics simulation of cosmic reionization and galaxy formation and their mutual impact, to redshift z < 6. With 40963 particles and cells in a 94 Mpc box, it is large enough to model global reionization and its feedback on galaxy formation while resolving all haloes above 108 M⊙. Using the same hybrid CPU–GPU code RAMSES–CUDATON as CoDa I in Ocvirk et al. (2016), CoDa II modified and re-calibrated the subgrid star formation algorithm, making reionization end earlier, at z ≳ 6, thereby better matching the observations of intergalactic Lyman α opacity from quasar spectra and electron-scattering optical depth from cosmic microwave background fluctuations. CoDa II predicts a UV continuum luminosity function in good agreement with observations of high-z galaxies, especially at z = 6. As in CoDa I, reionization feedback suppresses star formation in haloes below ∼2 × 109 M⊙, though suppression here is less severe, a possible consequence of modifying the star formation algorithm. Suppression is environment dependent, occurring earlier (later) in overdense (underdense) regions, in response to their local reionization times. Using a constrained realization of lambda cold dark matter constructed from galaxy survey data to reproduce the large-scale structure and major objects of the present-day Local Universe, CoDa II serves to model both global and local reionization. In CoDa II, the Milky Way and M31 appear as individual islands of reionization, i.e. they were not reionized by the progenitor of the Virgo cluster, or by nearby groups, or by each other.

OpenMP 4.5 compiler optimization for GPU offloading

Ability to efficiently offload computational workloads to graphic processing units (GPUs) is critical for the success of hybrid CPU–GPU architectures, such as the Summit and Sierra supercomputing systems. OpenMP 4.5 is a high-level programming model that enables the development of architecture- and accelerator-independent applications. This article describes aspects of the OpenMP implementation in the IBM XL C/C++ and XL Fortran OpenMP compilers that aid programmers to achieve performance objectives. This includes an interprocedural static analysis the XL optimizer uses to specialize code generation of the OpenMP <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">distribute parallel do</i> loop within the dynamic context of a target region, and other compiler optimizations designed to reduce the overhead of data transferred to an offloaded target region. We introduce the heuristic used at runtime to select optimal grid sizes for offloaded target team constructs. These tuned heuristics lead to an average improvement of 2× in the runtime of several target regions in the SPEC ACCEL V1.2 benchmark suite. In addition to performance enhancement, this article also presents an advanced diagnostic feature implemented in the XL Fortran compiler to aid in debugging OpenMP applications offloaded to accelerators.

Optimized co-scheduling of mixed-precision neural network accelerator for real-time multitasking applications

Abstract Neural networks are increasingly applied into real-time and embedded Artificial Intelligent (AI) systems like autonomous driving system. Such resource-constrained systems cannot support the execution of neural network based tasks due to their high execution overheads on general processors. Hence, we are approaching to design real-time AI applications on embedded systems with CPU and FPGA (Field Programmable Gate Array) coprocessors. We use dedicated FPGA to accelerate the neural network job and utilize CPU to process the rest jobs of real-time multitasking applications. We devise an Idle-Aware Earliest Deadline First policy to co-schedule the AI applications on hybrid CPU and FPGA coprocessors. Since the implementation of neural network job on FPGA accelerator with different precision configuration will result in different execution time and accuracy, we are also interested in the design optimization of real-time AI applications running on mixed-precision neural network accelerator, with the purpose of maximizing the accuracy related rewards of all applications subject to real-time related constraints. We address the problem as a multi-stage decision procedure, and propose an efficient dynamic programming approach with two pruning policies to reduce the intermediate searching states. Extensive experiments and real-life case evaluations demonstrate the efficiency of the proposed approaches.

Parallel reduction of four matrices to condensed form for a generalized matrix eigenvalue algorithm

The VZ algorithm proposed by Charles F. Van Loan (SIMA, 1975) attempts to solve the generalized type of matrix eigenvalue problem ACx = λBDx, where A, B ∈ Rn×m, C, D ∈ Rm×n, and m ≥ n, without forming products and inverses. Especially, this algorithm is suitable for solving the generalized singular value problem. Van Loan’s approach first reduces the matrices A, B, C, and D to a condensed form by the finite step initial reduction. The reduction finds orthogonal matrices Q, U, V, and Z, such that QAZ is upper Hessenberg, and QBV, ZTCU, and VTDU are upper triangular. In this initial reduction, A is reduced to upper Hessenberg form, while simultaneously preserving triangularity of other three matrices. This is done by Givens rotations, annihilating one by one element of A, and by generating three more rotations applied to other matrices per each annihilation. Such an algorithm is quite inefficient. In our work, we propose a blocked algorithm for the initial reduction, based on aggregated Givens rotations and matrix–matrix multiplications, which are applied in the outer loop updates. This algorithm has another level of blocking, exploited in the inner loop. Further, we also consider a variant of the algorithm in a hybrid CPU–GPU framework, where compute-intensive outer loop updates are performed on GPU, and can be overlapped with the reduction in the next step performed on CPU. On the other hand, application of a sequence of rotations in the inner loop is parallelized on CPU, with balanced operation count per thread. Since a large number of aggregated rotations are produced in every outer loop step, they are simultaneously accumulated before outer loop updates. These adjustments speed up original initial reduction considerably which is confirmed by numerical experiments, and the efficiency of the whole VZ algorithm is increased.

Accelerated CPU–GPUs implementations for quaternion polar harmonic transform of color images

Image moments are used to capture image features. Moments are successfully used in object descriptions, recognition, and other applications. However, image representation and recognition using quaternion moments are compute-intensive processes. This obstacle makes this type of moment unsuitable for real-time or large-scale applications, despite their high accuracy. In this work, we expose the challenges to parallelizing quaternion moment computations and transform the sequential algorithm to a parallel-friendly counterpart. We propose a set of parallel implementations for parallelizing quaternion moment image representations on different parallel architectures. The proposed implementations target multicore CPU-only, GPU-only, and hybrid CPU–GPUs (with multicore CPU and multi-GPU) environments. The loop mitigation technique is proposed to boost the level of parallelism in massively parallel environments, balance the parallel workload, and reduce both the space complexity and synchronization overhead of the proposed implementations. The loop mitigation technique could be applicable to other applications with similar loop imbalances. Finally, experiments are performed to evaluate the proposed implementations. Applying a moment order of 60 on a color image of 1024×1024 pixels, the proposed implementation achieved, on four P100 GPUs and a CPU with 16 cores, speedup of 257× and 277× over the baseline performance on a single Intel Xeon E5-2609 CPU core for image reconstruction and quaternion moment computation, respectively. In addition, 180× speedup is achieved for color image watermarking application.

Towards extreme scale dissipative particle dynamics simulations using multiple GPGPUs

A multi-GPGPU development for Mesoscale Simulations using the Dissipative Particle Dynamics method is presented. This distributed GPU acceleration development is an extension of the DL_MESO package to MPI+CUDA in order to exploit the computational power of the latest NVIDIA cards on hybrid CPU–GPU architectures. Details about the extensively applicable algorithm implementation and memory coalescing data structures are presented. The key algorithms’ optimizations for the nearest-neighbour list searching of particle pairs for short range forces, exchange of data and overlapping between computation and communications are also given. We have carried out strong and weak scaling performance analyses with up to 4096 GPUs. A two phase mixture separation test case with 1.8 billion particles has been run on the Piz Daint supercomputer from the Swiss National Supercomputer Center. With CUDA aware MPI, proper GPU affinity, communication and computation overlap optimizations for multi-GPU version, the final optimization results demonstrated more than 94% efficiency for weak scaling and more than 80% efficiency for strong scaling. As far as we know, this is the first report in the literature of DPD simulations being run on this large number of GPUs. The remaining challenges and future work are also discussed at the end of the paper.

Crossmodal attentive skill learner: learning in Atari and beyond with audio–video inputs

This paper introduces the Crossmodal Attentive Skill Learner (CASL), integrated with the recently-introduced Asynchronous Advantage Option-Critic architecture [Harb et al. in When waiting is not an option: learning options with a deliberation cost. arXiv preprint arXiv:1709.04571, 2017] to enable hierarchical reinforcement learning across multiple sensory inputs. Agents trained using our approach learn to attend to their various sensory modalities (e.g., audio, video) at the appropriate moments, thereby executing actions based on multiple sensory streams without reliance on supervisory data. We demonstrate empirically that the sensory attention mechanism anticipates and identifies useful latent features, while filtering irrelevant sensor modalities during execution. Further, we provide concrete examples in which the approach not only improves performance in a single task, but accelerates transfer to new tasks. We modify the Arcade Learning Environment [Bellemare et al. in J Artif Intell Res 47:253–279, 2013] to support audio queries (ALE-audio code available at https://github.com/shayegano/Arcade-Learning-Environment), and conduct evaluations of crossmodal learning in the Atari 2600 games H.E.R.O. and Amidar. Finally, building on the recent work of Babaeizadeh et al. [in: International conference on learning representations (ICLR), 2017], we open-source a fast hybrid CPU–GPU implementation of CASL (CASL code available at https://github.com/shayegano/CASL).