Assessment of Graphic Processing Units (GPUs) for Department of Defense (DoD) Digital Signal Processing (DSP) Applications

In this report we analyze the performance of the fast Fourier transform (FFT) on graphics hardware (the GPU), comparing it to the best-of-class CPU implementation FFTW. We describe the FFT, the architecture of the GPU, and how general-purpose computation is structured on the GPU.We then identify the factors that influence FFT performance and describe several experiments that compare these factors between the CPU and the GPU.We conclude that the overhead of transferring data and initiating GPU computation are substantially higher than on the CPU, and thus for latencycritical applications, the CPU is a superior choice. We show that the CPU implementation is limited by computation and the GPU implementation by GPU memory bandwidth and its lack of a writable cache. TheGPU is comparatively better suited for larger FFTs withmany FFTs computed in parallel in applications where FFT throughput is most important; on these applications GPU and CPU performance is roughly on par. We also demonstrate that adding additional computation to an application that includes the FFT, particularly computation that is GPU-friendly, puts the GPU at an advantage compared to the CPU. The future of desktop processing is parallel. The last few years have seen an explosion of single-chip commodity parallel architectures that promises to be the centerpieces of future computing platforms. Parallel hardware on the desktop—new multicore microprocessors, graphics processors, and stream processors—has the potential to greatly increase the computational power available to today’s computer users, with a resulting impact in computation domains such asmultimedia, entertainment, signal and image processing, and scientific computation. Several vendors have recently addressed this need for parallel computing: Intel andAMD are producingmulticore CPUs; the IBMCell processor delivers impressive performancewith its  parallel cores; and several stream processor startups, including Stream Processors Inc. and Clearspeed, are producing commercial parallel stream processors. None of these chips, however, have achieved market penetration to the degree of the graphics processor (GPU). Today’s GPU features massive arithmetic capability and memory bandwidth with superior performance and price-performance when compared to the CPU. For instance, the NVIDIA JohnD.Owens, Shubhabrata Sengupta, andDaniel Horn. “Assessment of Graphic ProcessingUnits (GPUs) for Department of Defense (DoD) Digital Signal Processing (DSP) Applications”. Technical Report ECE-CE-, Computer Engineering Research Laboratory, University of California, Davis, . http://www.ece. ucdavis.edu/cerl/techreports/-/