Title :
GPUs as high-performance random sources
Author :
Changwoo Pyo ; Gyungho Lee
Author_Institution :
Dept. of Comput. Eng., Hongik Univ., Seoul, South Korea
Abstract :
A non-algorithmic physical random bit generator is introduced, using a graphic processing unit (GPU) as a random source. The generator takes randomness from parallel threads executing atomic instructions. The generator in terms of pass rates from the National Institute of Standards and Technology (NIST)´s statistical test suite and bit generation throughput is evaluated. The generator compares favourably with pseudorandom bit generators known to have fine statistical properties. The test bed Nvidia´s GPU GTX 680 can generate random bits at the rate of hundreds of megabits per second. The cost-effectiveness of GPUs as random key generators can be increased, as GPUs are now readily available in most desktop computers, and even in smart phones.
Keywords :
graphics processing units; random number generation; statistical analysis; NIST; National Institute of Standards and Technology; Nvidia GPU GTX 680; atomic instructions; bit generation throughput evaluation; desktop computers; graphic processing unit; high-performance random sources; nonalgorithmic physical random bit generator; parallel threads; random key generators; smart phones; statistical test suite evaluation;
Journal_Title :
Electronics Letters
DOI :
10.1049/el.2013.4047