BROWSE
Detail
Conference Paper
A Study of CUDA Acceleration and Impact of Data Transfer Overhead in Heterogeneous Environment
Cited - time in 
Share 
Share
- Authors
- Fahian Ahmed, Saddam Quirem, Byeong Kil Lee, Bum Joo Shin, Duk Joo Son, Young Choon Woo, Wan Choi
- Issue Date
- 2012-02
- Citation
- International Symposium on High Performance Computer Architecture (HPCA) 2012, pp.16-19
- Language
- English
- Type
- Conference Paper
- Abstract
- Along with the introduction of many-core GPUs, there is widespread interest in using GPUs to accelerate non-graphics applications such as energy, bioinformatics, finance and several research areas. With a wide range of data sizes where the CPU has greater performance, it would be important that CUDA enabled programs properly select when to and not to utilize the GPU for acceleration. Algorithms that use dynamic programming like P7Viterbi algorithm of HMMER 3.0 (genetic application) show high parallelism in its code. Based on performance hotspot analysis, these parallel features were exploited through the use of CUDA and a GPGPU. The CUDA implementation of this algorithm being performed on the Tesla C1060 enabled a 10-15X speedup depending on the number of queries. In this paper, we focus on accelerating HMMER 3.0 -one of the genetic applications with GPUs as co-processors. Also we investigate the potential performance bottleneck in GPU-CPU environment with blowfish -a security application. Based on workload characterization and bottleneck analysis, we provide optimization methodologies to remove the bottleneck.
- KSP Keywords
- Bottleneck analysis, CUDA Acceleration, Data transfer, High parallelism, Hotspot analysis, Many-Core, Potential performance, Tesla C1060, Viterbi algorithm, co-processor, dynamic Programming