Xref: utzoo comp.lang.fortran:5698 comp.lang.c:40036 comp.lang.c++:14072 Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!uakari.primate.wisc.edu!samsung!umich!wsu-cs!pandora.cs.wayne.edu!kil From: kil@pandora.cs.wayne.edu (Joseph I. Landman) Newsgroups: comp.lang.fortran,comp.lang.c,comp.lang.c++ Subject: Re: Running long memory intensive programs! Message-ID: <1991Jun14.012310.11962@cs.wayne.edu> Date: 14 Jun 91 01:23:10 GMT References: <1991Jun12.203754.9356@noose.ecn.purdue.edu> <1991Jun13.162533.24651@afit.af.mil> Reply-To: kil@pandora.cs.wayne.edu (Joseph I. Landman) Organization: Wayne State University, Detroit Lines: 47 I am quite familiar with the use of 286/386 type machines, and the fortran compilers ms fortran 5.0, and the lahey f77-lem/16. I must state that for memory intensive/compute intensive work I would NOT recommend a strict 386/387 system. The benchmark speeds that have some applicability inidcate that they run about .25 MegaFlop for a 33Mhz system. But if you have this system and you cannot afford the funds to upgrade, then I would recommend the following course of action. Buy the Microsoft fortran (v5.1) and OS/2, the unbundled version. this option is inexpensive in comparison to getting unix for the 386 with a compiler, etc. OS/2 unbundled can be had for about 150$, and it will let ms fortran use ALL of available memory for its data objects. MS fortran will generate re-entrant multithreaded code that will run under OS/2 (it comes with os/2 specific libraries). Plus, this code will run/compile etc in the background if you wish. This is probably the least expensive option, and possibly the simplest. Others have suggested unix on the 386. Well Ive been using PC's since IBM introduced them, from measly little 8088 based machines (like the one I am typing on now), to speedster 486 machines with 16Mb memory, etc. Unix on these machines is horrible. Not the implementation, just the speed. If you are willing to spend the ridiculous amount of money on a new operating system that some of the people suggested, you may as well get better hardware instead. Such as an i860 based addin card for the 386 machine with a fortran compiler. I have seen ttthe torq, and i860 based workstation. with only one i860 processor, fortran flies at about 12 Mflops . The addin cards are simialr I understand. Thus for about the price of a new (slow) sparc 1, you can get the performance of an rs 6000 / 520. I personally do extremely compute intensive non vectorizable work that requires raw megafloppage. THe local IBM3090 is not fast enough for me, my short runs take about aCn hour, at midnight. I tried this program on my 286/10. Well I let the machine run over the weekend, and voila, monday morning, it was still chugging away..., not even through the 1st iteration of several hundred. I have used the lahey compiler f77-lem 16. It is a bit flaky on the 286/10, but I am very impressed by the speed and quality on the 386/387 in my thesis advisors office. If it were more reasonably priiced, I would have my own personal copy for home, but I have as of yet to see a reasonably priced fortran compiler for PC's. Joe Landman kil@pandora.cs.wayne.edu D