Item response theory (IRT) is a popular approach used for addressing large-scale statistical problems in psychometrics as well as in\nother fields. The fully Bayesian approach for estimating IRT models is usually memory and computationally expensive due to the\nlarge number of iterations. This limits the use of the procedure inmany applications. In an effort to overcome such restraint, previous\nstudies focused on utilizing the message passing interface (MPI) in a distributed memory-based Linux cluster to achieve certain\nspeedups. However, given the high data dependencies in a single Markov chain for IRT models, the communication overhead\nrapidly grows as the number of cluster nodes increases. This makes it difficult to further improve the performance under such\na parallel framework. This study aims to tackle the problem using massive core-based graphic processing units (GPU), which is\npractical, cost-effective, and convenient in actual applications.The performance comparisons among serial CPU, MPI, and compute\nunified device architecture (CUDA) programs demonstrate that the CUDA GPU approach has many advantages over the CPUbased\napproach and therefore is preferred.
Loading....