AbstractID: 8510 Title: Implementation of the Dose Planning Method (DPM) Monte Carlo code on a parallel processing network We have parallelized Dose Planning Method (DPM) Monte Carlo code, with the ultimate intent of routine radiotherapy dose calculations, on an Intel cluster and AMD Linux cluster at the University of Michigan Center for Advanced Computing. The Intel cluster consists of 50 dual processor, 800 MHz Pentium III nodes each with 1 GB RAM and the AMD cluster consists of 50 dual processor AMD Athlon XP 1800 MHz nodes each with 1 GB RAM. Parallelization was performed using the message passing interface (MPI) library for interprocessor communication. Uncorrelated random number streams were generated using the Scalable Parallel Psuedo-Random Number Generator (SPRNG) library. The appropriate subroutine within DPM was modified to incorporate SPRNG. Temporary buffers, assigned within DPM, are used by the master and slave processors to store dose and error values; this significantly reduces the memory requirement for dose calculations on individual processors. Initial evaluation shows a linear increase in computing speed with number of processors – currently we are using up to 32 processors. There is a noted departure from linearity in speed-up and efficiency at a larger number of processors because of inter-processor communication and synchronization. Calculations on the AMD cluster show that it is significantly faster than those on the Intel cluster due to the factor of two increase in processor speed. However, increasing the number of processors appears to reduce the efficiency on the AMD cluster faster than that on the Intel cluster. problem. Optimization of the MPI broadcast calls is expected to alleviate this