doherty@vax1.tcd.ie (07/21/90)
I am involved in setting up a micro-computer lab for general student use in a university. The intention is to have a 486 based server and workstations consisting of floppy drive 386Sxs all running under 3+Open. Each of the workstations will also have a 387 Sx math chip. The network will be ethernet bases and be bridged to the campus ethernet to permit TCP/IP access to other College machines. It has now been suggested that we should consider using workstations based on 386 DXs but without math chips to cut costs. The suggestion is that math chips can be added in a later stage when their cost will have dropped and that 386/25s will be a better long term investment. I would be grateful for any advice or comments on this project. Michael Doherty, Assistant Director, Computer Lab, University of Dublin, Trinity College, Dublin 2. Tel 353-1-772941 x 1751, Fax 353-1-792587
RFM@psuvm.psu.edu (07/22/90)
The August 1990 issue of PC RESOURCE (the LAST issue, **SOB**) has an article discussing coprocessors. Sets down the issues. Answers???? - Judgement call... If most student apps are text-based, I'd go with SXs. If apps are a lot of math programming and engineering CAD/CAM, the DX/25s might be as fast as SXs with coprocessors. If the DX/25s have cache memory, they'll probably be pretty fast without coprocessors. How much memory (RAM) comes installed? Whichever microprocessor, I'd make sure I had as much RAM as I could cram into the boxes. Network software eats up lots of base 640k. IMHO, Bob M., PSU-Harrisburg.
baer@uwovax.uwo.ca (07/22/90)
In article <90202.224428RFM@psuvm.psu.edu>, RFM@psuvm.psu.edu writes: > The August 1990 issue of PC RESOURCE (the LAST issue, **SOB**) has an article > discussing coprocessors. Sets down the issues. Answers???? - Judgement call... > If most student apps are text-based, I'd go with SXs. If apps are a lot of > math programming and engineering CAD/CAM, the DX/25s might be as fast as > SXs with coprocessors. If the DX/25s have cache memory, they'll probably > be pretty fast without coprocessors. --------------------------------------------------------------------- Be careful if the application is truly math-intensive. On a math-intensive application with relatively little I/O, I'm finding: a) a 386sx + 387sx combination at 16mHz. is about 1/2 as fast, perhaps a bit better, than a 386dx + 387dx combination at 20 mHz. (no cache) b) the 386sx + 387sx combination at 16mHz. is about twice as fast (perhaps a bit better) than a 386dx with cache at 25mHz. but without a math chip, c) the 386dx with cache without a math chip is a slight bit slower than a 286 + 287 at 10 Mhz. (287 running at 2/3 clock speed of course). Douglas Baer, University of Western Ontario, London, Canada N6A 5C2 Internet: BAER@UWO.CA Bitnet: BAER@UWOVAX
dmurdoch@watstat.uwaterloo.ca (Duncan Murdoch) (07/23/90)
In article <6499.26a985cd@uwovax.uwo.ca> baer@uwovax.uwo.ca writes: > c) the 386dx with >cache without a math chip is a slight bit slower than a 286 + 287 at 10 Mhz. >(287 running at 2/3 clock speed of course). This makes it sound as though the 20Mhz 386 would also be slower than my 8 Mhz 8086+8087 on math intensive calculations. Makes me feel a little less obsolete :-). Duncan Murdoch
RFM@psuvm.psu.edu (07/23/90)
WOW! D. Baer's data on performances of various combinations of 386&387 chips - sx vs dx, etc. - are eye-opening. Perhaps our Dublin colleague ought to go for 20-mhx 386dx plus 387 chips -- would cut costs a bit. I still stand by my other comment, that he needs to ensure lots of extended memory and software to access it in his networked 306 macines -- whatever they turn out to be.
brad@looking.on.ca (Brad Templeton) (07/24/90)
It's not an easy question... - 387s are expensive, that's true. - It is possible to write a very decent software math library for the 386, even in real mode, that uses the 32 bit registers. You can get acceptable FP performance for all but truly FP-intensive applications. Some programs now do this, like my own 3-2-1 Blastoff. - Most applications are still the stupid old DOS type, that do their software FP with an 8087 emulator written with 16 bit register instructions. S L O W. - A 486 is cheaper than a 386+387. If you plan for 387s from the start, the 486 makes more sense, particularly because it is a lot faster at FP. - The 486 may be cheaper, but the motherboards are not. This is strictly due to supply and demand, since in fact a 486 motherboard (with no cache, no 387 socket, only 25 mhz parts) should be a fair bit cheaper than a 386 board. Instead it costs twice as much. Within a year the price of these should drop. -- Brad Templeton, ClariNet Communications Corp. -- Waterloo, Ontario 519/884-7473
koziarz@halibut.nosc.mil (Walter A. Koziarz) (07/27/90)
The answer depends [obviously] on how much flaoting point math is really performed. As a point of comparison I offer the following -- 25MHz 80386DX - NO coproc runs *my* WHETSONE benchmark (8086 code) in 19 seconds. 10MHz 8088 - WITH 8087 (both @ 10MHz) runs same code in 6.7 seconds. Hope this has helped..... Walt K.
koziarz@halibut.nosc.mil (Walter A. Koziarz) (07/27/90)
In article <1990Jul22.202617.5437@maytag.waterloo.edu> dmurdoch@watstat.uwaterloo.ca (Duncan Murdoch) writes: >In article <6499.26a985cd@uwovax.uwo.ca> baer@uwovax.uwo.ca writes: >> c) the 386dx with >>cache without a math chip is a slight bit slower than a 286 + 287 at 10 Mhz. >>(287 running at 2/3 clock speed of course). > >This makes it sound as though the 20Mhz 386 would also be slower than my 8 Mhz >8086+8087 on math intensive calculations. Makes me feel a little less >obsolete :-). > It doesn't just *sound* slower; it *most likely IS* slower. My 10MHz 8088+10MHz 8087 combo (both running at 10MHz btw) is a little better than 3x faster than the '386DXs here at work (without '387DXs, obviously). If math is your major use of the machine you are *NOT* obsolete. Walt K.