We are intending to buy a powerful workstation for our daily processing of some seismic data. The process consists of doing numeric processing on data acquired on different seismic stations and by its nature the stations and their respective data and the processing on them are completely independent for the most part. I have been running some tests on different architectures. The code is written in openmp and the task is distributed on the number of cpu cores. On an Intel XEON with eight cores the time was reduced by almost a factor of 8, while on an AMD Opteron with 24 cores and almost comparable clock frequency to INTEL, it used almost 10 times longer to do the same process (alle 24 cores were employed)! When I reduced the "number of cores" to one on both machines, AMD was still slower by almost the same factor. The data set can be very large. The AMD has much larger physical memory available but the "cache size" is 4 times smaller then Intel. I need some advice on how these parameters affect the speed (especially the cache size in parallel programming context) so we would use the money on a suitable hardware. The platform is Linux.
Thanks in advance