Performance Analysis of GPU Accelerators with Realizable Utilization of Computational Density

Graphic Processing Units (GPUs) have become the accelerator of choice for improving the performance of many of the most demanding applications. While performance of these devices continues to improve generation after generation, reliability of these devices has not been studied rigorously. Several sources of errors can undermine the reliability of these devices, including radiation-induced transient faults, environmental perturbations, and process, temperature or voltage variations. In particular, transient faults in GPU execution have become a significant threat to high performance computing (HPC) and safety-critical applications. HPC systems experience transient faults every few tens of hours and the trend is expected to become worse. A key point in the study of transient faults and their effects on user programs is that some faults do not cause undesirable results in the affected programs. This is especially true for GPU applications. Past efforts to study and understand GPU vulnerability to transient faults have demonstrated reliability can vary greatly across different applications. A significant amount of resilience resides intrinsically in some GPU applications. Transient faults in these applications are not likely to affect the results they produce. Other applications are highly sensitive to transient faults and are likely to crash or produce incorrect results when they are affected by transient faults. While it is generally a good idea to protect the GPU hardware from transient faults, the penalties incurred in terms of performance, power and area are not always justifiable, depending on the applications utilizing the hardware resources. Understanding the relationship between the underlying program characteristics and their implications on vulnerability is crucial. The inherent resilience in applications should be carefully considered when making decisions about designing protection mechanisms to guard against nefarious transient faults. In this thesis, we focus on program characteristics that contribute to their vulnerability. We offer several methodologies that aim at alleviating the prohibitively expensive process of quantifying and estimating the vulnerability of GPU applications, as this is the first step toward improving the reliability of GPUs. Our analyses also demonstrate that beyond the variability between the vulnerabilities of different applications, the vulnerability of GPU applications also varies during their runtime, and present a phase behavior. This phase behavior opens new opportunities for not only more efficient vulnerability estimation, but also more efficient fault mitigation approaches. We demonstrate a methodology for application designers to reduce the cost of protection against transient faults.

Graphics Processing Unit (GPU) has emerged as a popular computing device to achieve Exa-scale performance in High-Performance Computing (HPC) applications. While the power-performance ratio is relatively high for a GPU, it still draws a significant amount of power during computation. In this paper, we propose a preliminary power prediction model which can be used by developers for building power-efficient GPU applications. Using this proposed work, developers can estimate the power consumption of a GPU application during implementation without having to execute it on actual hardware. Our model combines the information derived from static analysis of a CUDA program and a machine learning-based model. We have utilised decision tree technique to validate results across three different GPU architectures: Kepler, Maxwell and Volta. Observed <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> score value using the decision tree model is 0.8973 for Volta architecture.

Duplication with Comparison (DWC) is an effective software-level solution to improve the reliability of computing systems, including Graphics Processing Units (GPUs). DWC, however, introduces performance and energy consumption overheads that could be unacceptable for High-Performance Computing (HPC) or real-time safety-critical applications. In this work, we propose Reduced-Precision DWC (RP-DWC): an improvement over the traditional DWC approach that uses mixed-precision GPUs hardware resources to implement fault detection. We investigate, through both fault injection campaigns and accelerated neutron beam experiments, the impact of RPDWC onto performance, energy consumption, and its fault detection capabilites. We show that RP-DWC achieves on average 74% fault coverage (up to 86%) with very small overheads (0.1% time and 24% energy consumption overhead, in the best case).

Graphics Processing Units (GPUs) have become pivotal for modern high-performance computing (HPC) and artificial intelligence workloads due to their substantial computational prowess. However, this computational prowess comes at a cost, as GPUs consume vast amounts of power, presenting a challenge for high-end computing systems, including those aimed at achieving exascale computing capabilities. In response to the power efficiency problem, modern GPUs typically offer the ability to adjust clock frequencies and cap power consumption. However, the AMD Instinct MI100 takes a unique approach by introducing a set of predefined power profiles that internally manipulate clock frequencies to manage power. This study evaluates the effectiveness of these power profiles through a comparative analysis of various power and performance metrics. It indicates that, for most of the selected workloads and during significant portions of their execution, the GPU consumes power exceeding its specified Thermal Design Power (TDP). For instance, the GROMACS workload exceeded its TDP by one-third during almost half of its execution time. Furthermore, the study notes a significant increase in temperature reaching as high as 80°C. Moreover, DGEMM and STREAM workloads exhibit similar power consumption patterns, suggesting that the underlying power management scheme does not adapt power allocation based on the computational intensity of the workload. Thus, the study demonstrates that changing the power profile does not significantly impact crucial metrics such as performance, clock frequency, voltage, GPU utilization, or temperature. In summary, this research sheds light on the power dynamics of the AMD Instinct MI100 GPU, emphasizing the challenges associated with power, performance, and thermal management in HPC environments. The findings underscore the importance of fine-tuning power management strategies to enhance energy efficiency while maintaining optimal performance in GPUs.

With the rapid development of multimedia technologies and network communication, the parallel architecture such as the Graphic Processing Unit (GPU) is introduced in high-performance computing. But, how to program this GPU and how to obtain the best execution time remains usually an art. In this paper, a search study is performed on the Thread and the Block number that leads to a Prediction Unit of 64×64 (PU64) computation in the High Efficiency Video Coding (HEVC). It is proposed through the Compute Unified Device Architecture (CUDA). This method is described to optimize the GPU execution time. Experimental results show that the best Grid topology chosen to run the GPU kernel is obtained for 128 Block and 32 Thread. This proposed repartition gives the minimum GPU execution time compared to the CPU one, where the speed-up obtained here is around 50%.

Most Computational Fluid Dynamics (CFD) applications solve the Naiver-Stokes equations using various discretization methods like Finite Difference Method (FDM), Finite Element Method (FEM), Finite Volume Method (FVM), etc. The most compute-intensive process in CFD algorithms is the pressure Poisson solver. Solving the pressure Poisson equation requires the solution of a set of linear simultaneous equations. As the problem size/complexity increases, so does the effort and time invested in solving the equations. This study presents fast and robust iterative solvers and the acceleration of the same using the Graphics Processing Unit (GPU) architecture. Serial versions of the codes are implemented in the C language and the parallelization on GPU is achieved using Compute Unified Device Architecture (CUDA). For this study, a two-dimensional heat conduction problem is considered. FDM is used for discretization. Iterative solvers Conjugate Gradient and Multi Grid methods have been implemented in serial and parallel. The performance enhancement with single GPU and CPU system over a single CPU in terms of computing time is reported. It is seen that multigrid algorithms are superior in convergence and nearly 13 times speed-up has been obtained through GPU acceleration using CUDA, for a grid size of 2048 X 2048. The tests have been done on a cluster with Intel(R) Xeon(R) CPU E5-2670 CPU and NVIDIA K10 GPU.

The architecture of the latest Graphic Processing Unit (GPU) has surpassed the previous application-specific stream architecture. This has led to an architecture consisting of a number of uniform programmable units integrated on the same chip which facilitate the general-purpose computing beyond the graphic processing. With the multiple programmable units executing in parallel, the latest GPU shows superior performance. Furthermore, programmers can have a direct control on the GPU pipeline using easy-to-use parallel programming environments, whereas they had to rely on specific graphics API's in the past. These advances in hardware and software make General-Purpose GPU (GPGPU) computing widespread. In this paper, using the latest GPU and its software environment, we parallelize a computationally demanding financial application and optimize its performance. We also analyze the performance results compared with those obtained using CPU only. Experimental results show that GPU can achieve a superior performance, greater than 190x, compared with the CPU-only case.

Graphics processing units (GPUs) have shown increased popularity and play an important role as kind of coprocessor in heterogeneous co-processing environment. Heavily data parallel problems can be solved efficiently due to tens of thousands threads collaborative work in parallel GPU architecture. The achieved performance, therefore, depends on the capability of multiple threads in parallel collaboration. This paper, a static analytical kernel performance model (SAKP) was proposed to estimate the execution time of GPU kernel. Especially, a set of kernel and device features for target GPU is generated in the proposed model. We determine the performance limiting factors and generate an estimation of kernel execution time with this model. Matrix multiplication (MM) and histogram generation (HG) in NVIDIA GTX680 GPU card were performed to verify our proposed model and showed an absolute error in prediction less than 6.8%.

Multi-Agent Based Simulation (MABS) is used to study complex systems in many research domains. As the number of modeled agents is constantly growing, using General-Purpose Computing on Graphics Units (GPGPU) appears to be very promising as it allows to use the massively parallel architecture of the GPU (Graphics Processing Unit) to do High Performance Computing (HPC). However, this technology relies on a highly specialized architecture, implying a very specific programming approach. So, to benefit from GPU power, a MABS model need to be adapted to the GPU programming paradigm.

Machine virtualization and cloud computing environment have highlighted for last several years. This trend is based on the endeavor to enhance the utilization and reduce the ownership cost of machines. On the other hand, in aspect of high performance computing, graphics processing unit (GPU) has proved its capability for general purpose computing in many research areas. Evolving from traditional APIs such as the OpenGL and the Direct3D to program GPU as a graphics device, the CUDA of NIVDIA and the OpenCL provide more general programming environment for users. By supporting memory access model, interfaces to access GPUs directly and programming toolkits, users can perform parallel computation using the hundreds of GPU cores. In this paper, we propose a GPU virtualization mechanism to exploit GPU on virtualized cloud computing environment. Differently from the previous work which mostly reimplemented GPU programming APIs and virtual device drivers, our proposed mechanism uses the direct pass-through of PCI-E channel having GPU. The main limitation of previous approaches is virtualization overhead. Since they were focused on the sharing of GPU among virtual machines, they reimplemented GPU programming APIs at virtual machine monitor (VMM) level, and it incurred significant performance overhead. Moreover, if APIs are changed, they need to reengineer the most of APIs. In our approach, bypassing virtual machine monitor layer with negligible overhead, the mechanism can achieve similar computation performance to bare-metal system and is transparent to the GPU programming APIs.

In scientific computing, solving systems of linear equations is a very essential and widely used complex problem. Finding an optimal solution, using various algorithms, to these highly computation problems is very important. As the number of variables increases, the complexity of these problems also increases. These huge amounts of computations require High Performance Computing (HPC) systems, where efficient task distribution and accurate performance analysis are highly required on using these systems. In this paper, the computation performance using multicore architecture and Graphics Processing Units (GPUs) are compared with an analysis and expectations for the future of the new multicore systems. The number of cores in GPUs is extremely larger compared to the multicore architecture, while the memory overhead degrades the performance of GPUs dramatically. The expected scalability of the multicore system architecture tends to have better performance than GPUs.

List mode based image reconstruction is continuously gaining ground in PET imaging. The centerpiece of this reconstruction is the ray projector which unfortunately also represents the most time consuming part of the reconstruction process. Graphics processing units (GPU) represent today a cheap alternative to getting high computation power in medical imaging. However, only few GPU adapted methods have been proposed within the PET image reconstruction context, especially concerning the ray projector. In this work we propose and evaluate a new projector called DDA-ELL, which has the advantage of being applicable to both CPU and GPU architectures. The proposed projector was incorporated and evaluated within the 3D-OPLEM reconstruction algorithm implemented on both CPU and GPU architectures. Quantitative and qualitative performance evaluation was performed using simulated phantom and clinical datasets. We have shown that the proposed ray projector reconstructed images lead to improved noise and contrast characteristics compared to the widely used iterative Siddon approach. The speedup reached by the CPU version of our method compared to the iterative Siddon method was 11, while the GPU version was nearly 140 times faster.

In the last decade, Graphics Processing Units(GPUs) have gained an increasing popularity as accelerators for High Performance Computing (HPC) applications. Recent GPUs are not only powerful graphics engines but also highly threaded parallel computing processors that can achieve sustainable speedup as compared with CPUs. In this context, researchers try to exploit the capability of this architecture to solve difficult problems in many domains in science and engineering. In this article, we present recent advances on GPU Computing in Operations Research. We focus in particular on Integer Programming and Linear Programming.

When GeForce256 was issued in 1999, NVIDIA first proposes the concept of GPU (Graphics Processor Unit), and then a large number of complex application requirements make the whole industry flourishing. This paper first presents the development of GPU, and then briefly introduces and compares two popular development platform of GPU: CUDA (Compute Unified Device Architecture) C and OpenCL (Open Computing Language), and finally summarizes the successful applications of GPU computing in astronomy. And in the conclusions, it also presents that the problems about branch prediction capability, the larger cache and shared memory and the thread particle size may be solved in the further research about GPU computing.

The computational power and availability of graphics processing units (GPUs) and many integrated core (MIC) processors on high performance computing (HPC) systems is rapidly evolving. However, HPC applications need to be ported to take advantage of such hardware. This paper is a report on our experience of porting the MPI+OpenMP parallelised large-eddy simulation model (PALM) to multi-GPU as well as to MIC processor environments using OpenACC and OpenMP. PALM is written in Fortran, entails 140 kLOC and runs on HPC farms of up to 43,200 cores. The main porting challenges are the size and complexity of PALM, its inconsistent modularisation and no unit-tests. We report the methods used to identify performance issues as well as our experiences with state-of-the-art profiling tools. Moreover, we outline the required porting steps, describe the problems and bottlenecks we encountered and present separate performance tests for both architectures. We however, do not provide benchmark information.