Performance of Kepler GTX Titan GPUs and Xeon Phi System
Lattice Gauge Theory Research Center, CTP, and FPRD, Department of Physics and Astronomy, Seoul National University, Seoul, 151-747, South Korea
arXiv:1311.0590 [physics.comp-ph], (4 Nov 2013)
@article{2013arXiv1311.0590H,
author={Jeong, Hwancheol and Lee, Weonjong and Pak, Jeonghwan and Choi, Kwang-jong and Park, Sang-Hyun and Yoo, Jun-sik and Kim, Joo Hwan and Lee, Joungjin and Lee, Young Woo},
title={Performance of Kepler GTX Titan GPUs and Xeon Phi System},
journal={ArXiv e-prints},
archivePrefix={"arXiv"},
eprint={1311.0590},
primaryClass={"physics.comp-ph"},
keywords={Physics – Computational Physics, Computer Science – Distributed, Parallel, and Cluster Computing, High Energy Physics – Lattice},
year={2013},
month={nov}
}
NVIDIA’s new architecture, Kepler improves GPU’s performance significantly with the new streaming multiprocessor SMX. Along with the performance, NVIDIA has also introduced many new technologies such as direct parallelism, hyper-Q and GPU Direct with RDMA. Apart from other usual GPUs, NVIDIA also released another Kepler ‘GeForce’ GPU named GTX Titan. GeForce GTX Titan is not only good for gaming but also good for high performance computing with CUDA. Nevertheless, it is remarkably cheaper than Kepler Tesla GPUs. We investigate the performance of GTX Titan and find out how to optimize a CUDA code appropriately for it. Meanwhile, Intel has launched its new many integrated core (MIC) system, Xeon Phi. A Xeon Phi coprocessor could provide similar performance with NVIDIA Kepler GPUs theoretically but, in reality, it turns out that its performance is significantly inferior to GTX Titan.
November 6, 2013 by hgpu
