dbfunk@ICAEN.UIOWA.EDU (David B. Funk) (03/23/89)
We have a researcher who bought a DN3500 to run some chromatography modeling and analysis programs. These are computationaly intensive programs that he was running on a VAX and he wanted something faster. Our Apollo salesman sold him a DN3500 with a Weitek FPA. After doing some tests we found that his primary program runs twice as fast if he does NOT use the FPA. I ran diagnostics, the FPA works. It speeds up our benchmark programs but it slows down his. I am at a loss as to how to explain to this guy that his $2000 accelerator is slowing him down. Has anybody seen anything like this? Any ideas as to how to improve things? Here is some data from a small test case: OS: sr9.7.2; ftn revision 9.95 compiled with: -opt 4 -cpu 3000 program run cpu time: 32.157. real time: 36.084. compiled with: -opt 4 -cpu fpa1 program run cpu time: 61.558. real time: 65.949. OS: sr10.1; ftn Rev 10.6(46) compiled with: -opt 3 -cpu 3000 program run cpu time: 30.418. real time: 33.276. compiled with: -opt 3 -cpu fpa1 program run cpu time: 59.928. real time: 63.784. Dave Funk