| Abstract |
The Los Alamos National Laboratory Monte Carlo N-Particle radiation transport code, MCNP, is widely used around the world for many radiation protection and shielding applications. As a well-known standard it is also an excellent vehicle for assessing the relative performance of scientific computing platforms. Every three-to-four years a new version of MCNP is released internationally by the Radiation Safety Information Computational Center (RSICC) in Oak Ridge, Tennessee. For each of the past few releases, we have also done a timing study to assess the progress of scientific computing platforms and software. These quadrennial timing studies are valuable to the radiation protection and shielding community because (a) they are performed by a recognized scientific team, not a computer vendor, (b) they use an internationally recognized code for radiation protection and shielding calculations, (c) they are eminently reproducible since the code and the test problems are internationally distributed. Further, if one has a computer platform, operating system, or compiler not presented in our results, its performance is directly comparable to the ones we report because it can use the same code, data, and test problems as we used. Our results, using a single processor per platform, indicate that hardware advances during the past three years have improved performance by less than a factor of two and software improvements have had a marginal effect on performance. The most significant impacts on performance have resulted from developments in multiprocessing and multitasking. The other most significant advance in the last three years has been the accelerated improvements in personal computers. In the last timing study, the tested personal computer was approximately a factor of four slower that the fastest machine tested, a DEC Alphastation 500. In the present study, the fastest PC tested was less than a factor of two slower than the fastest platform, which is a Compaq (previously DEC) Alpha XP1000. |
