CUDA-for-clusters: a system for efficient execution of CUDA kernels on multi-core clusters

Rapid advancements in multi-core processor architectures coupled with low-cost, low-latency, high-bandwidth interconnects have made clusters of multi-core machines a common computing resource. Unfortunately, writing good parallel programs that efficiently utilize all the resources in such a cluster is still a major challenge. Various programming languages have been proposed as a solution to this problem, but are yet to be adopted widely to run performance-critical code mainly due to the relatively immature software framework and the effort involved in re-writing existing code in the new language. In this paper, we motivate and describe our initial study in exploring CUDA as a programming language for a cluster of multi-cores. We develop CUDA-For-Clusters (CFC), a framework that transparently orchestrates execution of CUDA kernels on a cluster of multi-core machines. The well-structured nature of a CUDA kernel, the growing popularity, support and stability of the CUDA software stack collectively make CUDA a good candidate to be considered as a programming language for a cluster. CFC uses a mixture of source-to-source compiler transformations, a work distribution runtime and a light-weight software distributed shared memory to manage parallel executions. Initial results on running several standard CUDA benchmark programs achieve impressive speedups of up to 7.5X on a cluster with 8 nodes, thereby opening up an interesting direction of research for further investigation.

Believe it or Not! Multicore CPUs can Match GPUs for FLOP-intensive Applications

In this work, we evaluate performance of a real-world image processing application that uses a cross-correlation algorithm to compare a given image with a reference one. The algorithm processes individual images represented as 2-dimensional matrices of single-precision floating-point values using O(n4) operations involving dot-products and additions. We implement this algorithm on a nVidia GTX 285 GPU using CUDA, and also parallelize it for the Intel Xeon (Nehalem) and IBM Power7 processors, using both manual and automatic techniques. Pthreads and OpenMP with SSE and VSX vector intrinsics are used for the manually parallelized version, while a state-of-the-art optimization framework based on the polyhedral model is used for automatic compiler parallelization and optimization. The performance of this algorithm on the nVidia GPU suffers from: (1) a smaller shared memory, (2) unaligned device memory access patterns, (3) expensive atomic operations, and (4) weaker single-thread performance. On commodity multi-core processors, the application dataset is small enough to fit in caches, and when parallelized using a combination of task and short-vector data parallelism (via SSE/VSX) or through fully automatic optimization from the compiler, the application matches or beats the performance of the GPU version. The primary reasons for better multi-core performance include larger and faster caches, higher clock frequency, higher on-chip memory bandwidth, and better compiler optimization and support for parallelization. The best performing versions on the Power7, Nehalem, and GTX 285 run in 1.02s, 1.82s, and 1.75s, respectively. These results conclusively demonstrate that, under certain conditions, it is possible for a FLOP-intensive structured application running on a multi-core processor to match or even beat the performance of an equivalent GPU version.

Fast image reconstruction for fluorescence microscopy

Real-time image reconstruction is essential for improving the temporal resolution of fluorescence microscopy. A number of unavoidable processes such as, optical aberration, noise and scattering degrade image quality, thereby making image reconstruction an ill-posed problem. Maximum likelihood is an attractive technique for data reconstruction especially when the problem is ill-posed. Iterative nature of the maximum likelihood technique eludes real-time imaging. Here we propose and demonstrate a compute unified device architecture (CUDA) based fast computing engine for real-time 3D fluorescence imaging. A maximum performance boost of 210x is reported. Easy availability of powerful computing engines is a boon and may accelerate to realize real-time 3D fluorescence imaging. Copyright 2012 Author(s). This article is distributed under a Creative Commons Attribution 3.0 Unported License. http://dx.doi.org/10.1063/1.4754604]

Improving GPGPU Concurrency with Elastic Kernels

Each new generation of GPUs vastly increases the resources available to GPGPU programs. GPU programming models (like CUDA) were designed to scale to use these resources. However, we find that CUDA programs actually do not scale to utilize all available resources, with over 30% of resources going unused on average for programs of the Parboil2 suite that we used in our work. Current GPUs therefore allow concurrent execution of kernels to improve utilization. In this work, we study concurrent execution of GPU kernels using multiprogram workloads on current NVIDIA Fermi GPUs. On two-program workloads from the Parboil2 benchmark suite we find concurrent execution is often no better than serialized execution. We identify that the lack of control over resource allocation to kernels is a major serialization bottleneck. We propose transformations that convert CUDA kernels into elastic kernels which permit fine-grained control over their resource usage. We then propose several elastic-kernel aware concurrency policies that offer significantly better performance and concurrency compared to the current CUDA policy. We evaluate our proposals on real hardware using multiprogrammed workloads constructed from benchmarks in the Parboil 2 suite. On average, our proposals increase system throughput (STP) by 1.21x and improve the average normalized turnaround time (ANTT) by 3.73x for two-program workloads when compared to the current CUDA concurrency implementation.

3-D GPU Based Real Time Diffuse Optical Tomographic System

3-Dimensional Diffuse Optical Tomographic (3-D DOT) image reconstruction algorithm is computationally complex and requires excessive matrix computations and thus hampers reconstruction in real time. In this paper, we present near real time 3D DOT image reconstruction that is based on Broyden approach for updating Jacobian matrix. The Broyden method simplifies the algorithm by avoiding re-computation of the Jacobian matrix in each iteration. We have developed CPU and heterogeneous CPU/GPU code for 3D DOT image reconstruction in C and MatLab programming platform. We have used Compute Unified Device Architecture (CUDA) programming framework and CUDA linear algebra library (CULA) to utilize the massively parallel computational power of GPUs (NVIDIA Tesla K20c). The computation time achieved for C program based implementation for a CPU/GPU system for 3 planes measurement and FEM mesh size of 19172 tetrahedral elements is 806 milliseconds for an iteration.

Real-time maximum a-posteriori image reconstruction for fluorescence microscopy

Rapid reconstruction of multidimensional image is crucial for enabling real-time 3D fluorescence imaging. This becomes a key factor for imaging rapidly occurring events in the cellular environment. To facilitate real-time imaging, we have developed a graphics processing unit (GPU) based real-time maximum a-posteriori (MAP) image reconstruction system. The parallel processing capability of GPU device that consists of a large number of tiny processing cores and the adaptability of image reconstruction algorithm to parallel processing (that employ multiple independent computing modules called threads) results in high temporal resolution. Moreover, the proposed quadratic potential based MAP algorithm effectively deconvolves the images as well as suppresses the noise. The multi-node multi-threaded GPU and the Compute Unified Device Architecture (CUDA) efficiently execute the iterative image reconstruction algorithm that is similar to 200-fold faster (for large dataset) when compared to existing CPU based systems. (C) 2015 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution 3.0 Unported License.