Performance of LS-DYNA with Double Precision on Linux and Windows CCS

Although the majority of LS-DYNA jobs are simulated with single-precision arithmetic, the more accurate and robust double-precision arithmetic has to be used for certain situations. Currently, most public LS-DYNA benchmarks are with single precision and few with double precision. As a result, users often try to extrapolate the performance of LS-DYNA with double precision from that with single precision. This paper shows that such extrapolations are often misleading. Furthermore, a comparative performance study on LS-DYNA with double precision, using all of the industry standard processors—Intel Xeon, Intel Itanium, and AMD Opteron—and the two major operating systems—Linux and Windows CCS—is presented to provide users with information for choosing the right configuration to run LS-DYNA with double precision.