Trends Of CPU, GPU and FPGA for high-performance computing

Floating-point computing with more than one TFLOP of peak performance is already a reality in recent Field-Programmable Gate Arrays (FPGA). General-Purpose Graphics Processing Units (GPGPU) and recent many-core CPUs have also taken advantage of the recent technological innovations in integrated circuit (IC) design and had also dramatically improved their peak performances. In this paper, we compare the trends of these computing architectures for high-performance computing and survey these platforms in the execution of algorithms belonging to different scientific application domains. Trends in peak performance, power consumption and sustained performances, for particular applications, show that FPGAs are increasing the gap to GPUs and many-core CPUs moving them away from high-performance computing with intensive floating-point calculations. FPGAs become competitive for custom floating-point or fixed-point representations, for smaller input sizes of certain algorithms, for combinational logic problems and parallel map-reduce problems. © 2014 Technical University of Munich (TUM).

GPU implementation of a hyperspectral coded aperture algorithm for compressive sensing

This paper presents a new parallel implementation of a previously hyperspectral coded aperture (HYCA) algorithm for compressive sensing on graphics processing units (GPUs). HYCA method combines the ideas of spectral unmixing and compressive sensing exploiting the high spatial correlation that can be observed in the data and the generally low number of endmembers needed in order to explain the data. The proposed implementation exploits the GPU architecture at low level, thus taking full advantage of the computational power of GPUs using shared memory and coalesced accesses to memory. The proposed algorithm is evaluated not only in terms of reconstruction error but also in terms of computational performance using two different GPU architectures by NVIDIA: GeForce GTX 590 and GeForce GTX TITAN. Experimental results using real data reveals signficant speedups up with regards to serial implementation.

GPU implementation of the simplex identification via split augmented Lagrangian

Hyperspectral imaging can be used for object detection and for discriminating between different objects based on their spectral characteristics. One of the main problems of hyperspectral data analysis is the presence of mixed pixels, due to the low spatial resolution of such images. This means that several spectrally pure signatures (endmembers) are combined into the same mixed pixel. Linear spectral unmixing follows an unsupervised approach which aims at inferring pure spectral signatures and their material fractions at each pixel of the scene. The huge data volumes acquired by such sensors put stringent requirements on processing and unmixing methods. This paper proposes an efficient implementation of a unsupervised linear unmixing method on GPUs using CUDA. The method finds the smallest simplex by solving a sequence of nonsmooth convex subproblems using variable splitting to obtain a constraint formulation, and then applying an augmented Lagrangian technique. The parallel implementation of SISAL presented in this work exploits the GPU architecture at low level, using shared memory and coalesced accesses to memory. The results herein presented indicate that the GPU implementation can significantly accelerate the method's execution over big datasets while maintaining the methods accuracy.

Parallel GPU architecture for hyperspectral unmixing based on augmented Lagrangian method

Hyperspectral imaging has become one of the main topics in remote sensing applications, which comprise hundreds of spectral bands at different (almost contiguous) wavelength channels over the same area generating large data volumes comprising several GBs per flight. This high spectral resolution can be used for object detection and for discriminate between different objects based on their spectral characteristics. One of the main problems involved in hyperspectral analysis is the presence of mixed pixels, which arise when the spacial resolution of the sensor is not able to separate spectrally distinct materials. Spectral unmixing is one of the most important task for hyperspectral data exploitation. However, the unmixing algorithms can be computationally very expensive, and even high power consuming, which compromises the use in applications under on-board constraints. In recent years, graphics processing units (GPUs) have evolved into highly parallel and programmable systems. Specifically, several hyperspectral imaging algorithms have shown to be able to benefit from this hardware taking advantage of the extremely high floating-point processing performance, compact size, huge memory bandwidth, and relatively low cost of these units, which make them appealing for onboard data processing. In this paper, we propose a parallel implementation of an augmented Lagragian based method for unsupervised hyperspectral linear unmixing on GPUs using CUDA. The method called simplex identification via split augmented Lagrangian (SISAL) aims to identify the endmembers of a scene, i.e., is able to unmix hyperspectral data sets in which the pure pixel assumption is violated. The efficient implementation of SISAL method presented in this work exploits the GPU architecture at low level, using shared memory and coalesced accesses to memory.

Parallel hyperspectral compressive sensing method on GPU

Remote hyperspectral sensors collect large amounts of data per flight usually with low spatial resolution. It is known that the bandwidth connection between the satellite/airborne platform and the ground station is reduced, thus a compression onboard method is desirable to reduce the amount of data to be transmitted. This paper presents a parallel implementation of an compressive sensing method, called parallel hyperspectral coded aperture (P-HYCA), for graphics processing units (GPU) using the compute unified device architecture (CUDA). This method takes into account two main properties of hyperspectral dataset, namely the high correlation existing among the spectral bands and the generally low number of endmembers needed to explain the data, which largely reduces the number of measurements necessary to correctly reconstruct the original data. Experimental results conducted using synthetic and real hyperspectral datasets on two different GPU architectures by NVIDIA: GeForce GTX 590 and GeForce GTX TITAN, reveal that the use of GPUs can provide real-time compressive sensing performance. The achieved speedup is up to 20 times when compared with the processing time of HYCA running on one core of the Intel i7-2600 CPU (3.4GHz), with 16 Gbyte memory.

Parallel hyperspectral unmixing method on GPU

Many Hyperspectral imagery applications require a response in real time or near-real time. To meet this requirement this paper proposes a parallel unmixing method developed for graphics processing units (GPU). This method is based on the vertex component analysis (VCA), which is a geometrical based method highly parallelizable. VCA is a very fast and accurate method that extracts endmember signatures from large hyperspectral datasets without the use of any a priori knowledge about the constituent spectra. Experimental results obtained for simulated and real hyperspectral datasets reveal considerable acceleration factors, up to 24 times.

Aceleración de Algoritmos en Procesadores Multicore, GPGPU y FPGA. Algorithm Acceleration with Multicore, GPGPU and FPGA Processors.

El avance en la potencia de cómputo en nuestros días viene dado por la paralelización del procesamiento, dadas las características que disponen las nuevas arquitecturas de hardware. Utilizar convenientemente este hardware impacta en la aceleración de los algoritmos en ejecución (programas). Sin embargo, convertir de forma adecuada el algoritmo en su forma paralela es complejo, y a su vez, esta forma, es específica para cada tipo de hardware paralelo. En la actualidad los procesadores de uso general más comunes son los multicore, procesadores paralelos, también denominados Symmetric Multi-Processors (SMP). Hoy en día es difícil hallar un procesador para computadoras de escritorio que no tengan algún tipo de paralelismo del caracterizado por los SMP, siendo la tendencia de desarrollo, que cada día nos encontremos con procesadores con mayor numero de cores disponibles. Por otro lado, los dispositivos de procesamiento de video (Graphics Processor Units - GPU), a su vez, han ido desarrollando su potencia de cómputo por medio de disponer de múltiples unidades de procesamiento dentro de su composición electrónica, a tal punto que en la actualidad no es difícil encontrar placas de GPU con capacidad de 200 a 400 hilos de procesamiento paralelo. Estos procesadores son muy veloces y específicos para la tarea que fueron desarrollados, principalmente el procesamiento de video. Sin embargo, como este tipo de procesadores tiene muchos puntos en común con el procesamiento científico, estos dispositivos han ido reorientándose con el nombre de General Processing Graphics Processor Unit (GPGPU). A diferencia de los procesadores SMP señalados anteriormente, las GPGPU no son de propósito general y tienen sus complicaciones para uso general debido al límite en la cantidad de memoria que cada placa puede disponer y al tipo de procesamiento paralelo que debe realizar para poder ser productiva su utilización. Los dispositivos de lógica programable, FPGA, son dispositivos capaces de realizar grandes cantidades de operaciones en paralelo, por lo que pueden ser usados para la implementación de algoritmos específicos, aprovechando el paralelismo que estas ofrecen. Su inconveniente viene derivado de la complejidad para la programación y el testing del algoritmo instanciado en el dispositivo. Ante esta diversidad de procesadores paralelos, el objetivo de nuestro trabajo está enfocado en analizar las características especificas que cada uno de estos tienen, y su impacto en la estructura de los algoritmos para que su utilización pueda obtener rendimientos de procesamiento acordes al número de recursos utilizados y combinarlos de forma tal que su complementación sea benéfica. Específicamente, partiendo desde las características del hardware, determinar las propiedades que el algoritmo paralelo debe tener para poder ser acelerado. Las características de los algoritmos paralelos determinará a su vez cuál de estos nuevos tipos de hardware son los mas adecuados para su instanciación. En particular serán tenidos en cuenta el nivel de dependencia de datos, la necesidad de realizar sincronizaciones durante el procesamiento paralelo, el tamaño de datos a procesar y la complejidad de la programación paralela en cada tipo de hardware. Today´s advances in high-performance computing are driven by parallel processing capabilities of available hardware architectures. These architectures enable the acceleration of algorithms when thes ealgorithms are properly parallelized and exploit the specific processing power of the underneath architecture. Most current processors are targeted for general pruposes and integrate several processor cores on a single chip, resulting in what is known as a Symmetric Multiprocessing (SMP) unit. Nowadays even desktop computers make use of multicore processors. Meanwhile, the industry trend is to increase the number of integrated rocessor cores as technology matures. On the other hand, Graphics Processor Units (GPU), originally designed to handle only video processing, have emerged as interesting alternatives to implement algorithm acceleration. Current available GPUs are able to implement from 200 to 400 threads for parallel processing. Scientific computing can be implemented in these hardware thanks to the programability of new GPUs that have been denoted as General Processing Graphics Processor Units (GPGPU).However, GPGPU offer little memory with respect to that available for general-prupose processors; thus, the implementation of algorithms need to be addressed carefully. Finally, Field Programmable Gate Arrays (FPGA) are programmable devices which can implement hardware logic with low latency, high parallelism and deep pipelines. Thes devices can be used to implement specific algorithms that need to run at very high speeds. However, their programmability is harder that software approaches and debugging is typically time-consuming. In this context where several alternatives for speeding up algorithms are available, our work aims at determining the main features of thes architectures and developing the required know-how to accelerate algorithm execution on them. We look at identifying those algorithms that may fit better on a given architecture as well as compleme

Heterogeneous multicore systems for signal processing

This thesis explores the capabilities of heterogeneous multi-core systems, based on multiple Graphics Processing Units (GPUs) in a standard desktop framework. Multi-GPU accelerated desk side computers are an appealing alternative to other high performance computing (HPC) systems: being composed of commodity hardware components fabricated in large quantities, their price-performance ratio is unparalleled in the world of high performance computing. Essentially bringing “supercomputing to the masses”, this opens up new possibilities for application fields where investing in HPC resources had been considered unfeasible before. One of these is the field of bioelectrical imaging, a class of medical imaging technologies that occupy a low-cost niche next to million-dollar systems like functional Magnetic Resonance Imaging (fMRI). In the scope of this work, several computational challenges encountered in bioelectrical imaging are tackled with this new kind of computing resource, striving to help these methods approach their true potential. Specifically, the following main contributions were made: Firstly, a novel dual-GPU implementation of parallel triangular matrix inversion (TMI) is presented, addressing an crucial kernel in computation of multi-mesh head models of encephalographic (EEG) source localization. This includes not only a highly efficient implementation of the routine itself achieving excellent speedups versus an optimized CPU implementation, but also a novel GPU-friendly compressed storage scheme for triangular matrices. Secondly, a scalable multi-GPU solver for non-hermitian linear systems was implemented. It is integrated into a simulation environment for electrical impedance tomography (EIT) that requires frequent solution of complex systems with millions of unknowns, a task that this solution can perform within seconds. In terms of computational throughput, it outperforms not only an highly optimized multi-CPU reference, but related GPU-based work as well. Finally, a GPU-accelerated graphical EEG real-time source localization software was implemented. Thanks to acceleration, it can meet real-time requirements in unpreceeded anatomical detail running more complex localization algorithms. Additionally, a novel implementation to extract anatomical priors from static Magnetic Resonance (MR) scansions has been included.

Heterogeneous Multi-core Architectures for High Performance Computing

This thesis deals with heterogeneous architectures in standard workstations. Heterogeneous architectures represent an appealing alternative to traditional supercomputers because they are based on commodity components fabricated in large quantities. Hence their price-performance ratio is unparalleled in the world of high performance computing (HPC). In particular, different aspects related to the performance and consumption of heterogeneous architectures have been explored. The thesis initially focuses on an efficient implementation of a parallel application, where the execution time is dominated by an high number of floating point instructions. Then the thesis touches the central problem of efficient management of power peaks in heterogeneous computing systems. Finally it discusses a memory-bounded problem, where the execution time is dominated by the memory latency. Specifically, the following main contributions have been carried out: A novel framework for the design and analysis of solar field for Central Receiver Systems (CRS) has been developed. The implementation based on desktop workstation equipped with multiple Graphics Processing Units (GPUs) is motivated by the need to have an accurate and fast simulation environment for studying mirror imperfection and non-planar geometries. Secondly, a power-aware scheduling algorithm on heterogeneous CPU-GPU architectures, based on an efficient distribution of the computing workload to the resources, has been realized. The scheduler manages the resources of several computing nodes with a view to reducing the peak power. The two main contributions of this work follow: the approach reduces the supply cost due to high peak power whilst having negligible impact on the parallelism of computational nodes. from another point of view the developed model allows designer to increase the number of cores without increasing the capacity of the power supply unit. Finally, an implementation for efficient graph exploration on reconfigurable architectures is presented. The purpose is to accelerate graph exploration, reducing the number of random memory accesses.

GPU-based Ray Tracing of Dynamic Scenes

Interactive ray tracing of non-trivial scenes is just becoming feasible on single graphics processing units (GPU). Recent work in this area focuses on building effective acceleration structures, which work well under the constraints of current GPUs. Most approaches are targeted at static scenes and only allow navigation in the virtual scene. So far support for dynamic scenes has not been considered for GPU implementations. We have developed a GPU-based ray tracing system for dynamic scenes consisting of a set of individual objects. Each object may independently move around, but its geometry and topology are static.

Particle methods parallel implementations by GP-GPU strategies

This paper outlines the problems found in the parallelization of SPH (Smoothed Particle Hydrodynamics) algorithms using Graphics Processing Units. Different results of some parallel GPU implementations in terms of the speed-up and the scalability compared to the CPU sequential codes are shown. The most problematic stage in the GPU-SPH algorithms is the one responsible for locating neighboring particles and building the vectors where this information is stored, since these specific algorithms raise many dificulties for a data-level parallelization. Because of the fact that the neighbor location using linked lists does not show enough data-level parallelism, two new approaches have been pro- posed to minimize bank conflicts in the writing and subsequent reading of the neighbor lists. The first strategy proposes an efficient coordination between CPU-GPU, using GPU algorithms for those stages that allow a straight forward parallelization, and sequential CPU algorithms for those instructions that involve some kind of vector reduction. This coordination provides a relatively orderly reading of the neighbor lists in the interactions stage, achieving a speed-up factor of x47 in this stage. However, since the construction of the neighbor lists is quite expensive, it is achieved an overall speed-up of x41. The second strategy seeks to maximize the use of the GPU in the neighbor's location process by executing a specific vector sorting algorithm that allows some data-level parallelism. Al- though this strategy has succeeded in improving the speed-up on the stage of neighboring location, the global speed-up on the interactions stage falls, due to inefficient reading of the neighbor vectors. Some changes to these strategies are proposed, aimed at maximizing the computational load of the GPU and using the GPU texture-units, in order to reach the maximum speed-up for such codes. Different practical applications have been added to the mentioned GPU codes. First, the classical dam-break problem is studied. Second, the wave impact of the sloshing fluid contained in LNG vessel tanks is also simulated as a practical example of particle methods

Optimal Piecewise Linear Function Approximation for GPU-based Applications

Many computer vision and human-computer interaction applications developed in recent years need evaluating complex and continuous mathematical functions as an essential step toward proper operation. However, rigorous evaluation of this kind of functions often implies a very high computational cost, unacceptable in real-time applications. To alleviate this problem, functions are commonly approximated by simpler piecewise-polynomial representations. Following this idea, we propose a novel, efficient, and practical technique to evaluate complex and continuous functions using a nearly optimal design of two types of piecewise linear approximations in the case of a large budget of evaluation subintervals. To this end, we develop a thorough error analysis that yields asymptotically tight bounds to accurately quantify the approximation performance of both representations. It provides an improvement upon previous error estimates and allows the user to control the trade-off between the approximation error and the number of evaluation subintervals. To guarantee real-time operation, the method is suitable for, but not limited to, an efficient implementation in modern Graphics Processing Units (GPUs), where it outperforms previous alternative approaches by exploiting the fixed-function interpolation routines present in their texture units. The proposed technique is a perfect match for any application requiring the evaluation of continuous functions, we have measured in detail its quality and efficiency on several functions, and, in particular, the Gaussian function because it is extensively used in many areas of computer vision and cybernetics, and it is expensive to evaluate.

3D model reconstruction using neural gas accelerated on GPU

In this work, we propose the use of the neural gas (NG), a neural network that uses an unsupervised Competitive Hebbian Learning (CHL) rule, to develop a reverse engineering process. This is a simple and accurate method to reconstruct objects from point clouds obtained from multiple overlapping views using low-cost sensors. In contrast to other methods that may need several stages that include downsampling, noise filtering and many other tasks, the NG automatically obtains the 3D model of the scanned objects. To demonstrate the validity of our proposal we tested our method with several models and performed a study of the neural network parameterization computing the quality of representation and also comparing results with other neural methods like growing neural gas and Kohonen maps or classical methods like Voxel Grid. We also reconstructed models acquired by low cost sensors that can be used in virtual and augmented reality environments for redesign or manipulation purposes. Since the NG algorithm has a strong computational cost we propose its acceleration. We have redesigned and implemented the NG learning algorithm to fit it onto Graphics Processing Units using CUDA. A speed-up of 180× faster is obtained compared to the sequential CPU version.

Aceleración de Algoritmos en Procesadores Multicore, GPGPU y FPGA

El avance en la potencia de cómputo en nuestros días viene dado por la paralelización del procesamiento, dadas las características que disponen las nuevas arquitecturas de hardware. Utilizar convenientemente este hardware impacta en la aceleración de los algoritmos en ejecución (programas). Sin embargo, convertir de forma adecuada el algoritmo en su forma paralela es complejo, y a su vez, esta forma, es específica para cada tipo de hardware paralelo. En la actualidad los procesadores de uso general más comunes son los multicore, procesadores paralelos, también denominados Symmetric Multi-Processors (SMP). Hoy en día es difícil hallar un procesador para computadoras de escritorio que no tengan algún tipo de paralelismo del caracterizado por los SMP, siendo la tendencia de desarrollo, que cada día nos encontremos con procesadores con mayor numero de cores disponibles. Por otro lado, los dispositivos de procesamiento de video (Graphics Processor Units - GPU), a su vez, han ido desarrollando su potencia de cómputo por medio de disponer de múltiples unidades de procesamiento dentro de su composición electrónica, a tal punto que en la actualidad no es difícil encontrar placas de GPU con capacidad de 200 a 400 hilos de procesamiento paralelo. Estos procesadores son muy veloces y específicos para la tarea que fueron desarrollados, principalmente el procesamiento de video. Sin embargo, como este tipo de procesadores tiene muchos puntos en común con el procesamiento científico, estos dispositivos han ido reorientándose con el nombre de General Processing Graphics Processor Unit (GPGPU). A diferencia de los procesadores SMP señalados anteriormente, las GPGPU no son de propósito general y tienen sus complicaciones para uso general debido al límite en la cantidad de memoria que cada placa puede disponer y al tipo de procesamiento paralelo que debe realizar para poder ser productiva su utilización. Los dispositivos de lógica programable, FPGA, son dispositivos capaces de realizar grandes cantidades de operaciones en paralelo, por lo que pueden ser usados para la implementación de algoritmos específicos, aprovechando el paralelismo que estas ofrecen. Su inconveniente viene derivado de la complejidad para la programación y el testing del algoritmo instanciado en el dispositivo. Ante esta diversidad de procesadores paralelos, el objetivo de nuestro trabajo está enfocado en analizar las características especificas que cada uno de estos tienen, y su impacto en la estructura de los algoritmos para que su utilización pueda obtener rendimientos de procesamiento acordes al número de recursos utilizados y combinarlos de forma tal que su complementación sea benéfica. Específicamente, partiendo desde las características del hardware, determinar las propiedades que el algoritmo paralelo debe tener para poder ser acelerado. Las características de los algoritmos paralelos determinará a su vez cuál de estos nuevos tipos de hardware son los mas adecuados para su instanciación. En particular serán tenidos en cuenta el nivel de dependencia de datos, la necesidad de realizar sincronizaciones durante el procesamiento paralelo, el tamaño de datos a procesar y la complejidad de la programación paralela en cada tipo de hardware.

Procesamiento de imágenes cerebrales en GPU

Abstract: As time has passed, the general purpose programming paradigm has evolved, producing different hardware architectures whose characteristics differ widely. In this work, we are going to demonstrate, through different applications belonging to the field of Image Processing, the existing difference between three Nvidia hardware platforms: two of them belong to the GeForce graphics cards series, the GTX 480 and the GTX 980 and one of the low consumption platforms which purpose is to allow the execution of embedded applications as well as providing an extreme efficiency: the Jetson TK1. With respect to the test applications we will use five examples from Nvidia CUDA Samples. These applications are directly related to Image Processing, as the algorithms they use are similar to those from the field of medical image registration. After the tests, it will be proven that GTX 980 is both the device with the highest computational power and the one that has greater consumption, it will be seen that Jetson TK1 is the most efficient platform, it will be shown that GTX 480 produces more heat than the others and we will learn other effects produced by the existing difference between the architecture of the devices.