gengenba@unipas.fmi.uni-passau.de (Michael Gengenbach) (06/09/89)
Hi folks, I know, that this question could mean war, but we really need some decision support. We are the Bavarian AI Research Center (FORWIS) and we have to buy our first computer equipment, which is not an easy job :-). We will buy some 40 UNIX-Workstations und about 3 large fileservers. Now there are two favorized Workstations: - DECStation 3100 - SPARCstation 1 or SPARCstation 330 As we do not have much knowledge in RISC-technology, we have a couple of questions: - which RISC-architecture (MIPS, SPARC) is more common in the U.S.? - what do you think about RISC vs. CISC in performance of AI-applications (Lisp, Prolog, Databases, image-processing), i.e. symbolic computation? - can one trust in MIPS-ratings (14 for DECStation, 12.5 SPARCstation 1)? - how about the software availability, especially AI software? If you're going to answer, please do not flame! Keep technical! Thank you in advance Michael -- Michael Gengenbach \/\/\/\/\/ gengenbach@unipas.fmi.uni-passau.de AI Research Center \/\/\/\/ gengenbach%unipas.fmi.uni-passau.de@unido.BITNET University of Passau /\/\/\/\ gengenbach%unipas.fmi.uni-passau.de@relay.CS.NET West Germany /\/\/\/\/\ Phone: +49 851/509-533
cyrus@pprg.unm.edu (Tait Cyrus) (06/10/89)
In article <486@unipas.fmi.uni-passau.de> gengenba@unipas.fmi.uni-passau.de (Michael Gengenbach) writes: >Hi folks, > >I know, that this question could mean war, but we really need some >decision support. We are the Bavarian AI Research Center (FORWIS) >and we have to buy our first computer equipment, which is not an >easy job :-). Like your first time through a toy store :^) >We will buy some 40 UNIX-Workstations und about 3 large fileservers. >Now there are two favorized Workstations: > > - DECStation 3100 > - SPARCstation 1 or SPARCstation 330 > >As we do not have much knowledge in RISC-technology, we have a couple >of questions: > > - which RISC-architecture (MIPS, SPARC) is more common in the U.S.? That depends on whom you talk to. > - what do you think about RISC vs. CISC in performance of > AI-applications (Lisp, Prolog, Databases, image-processing), > i.e. symbolic computation? RISC has higher performance, though some of the newer CISCs (like the new 64/128 bit one from Intel) are going to be some competition. > - can one trust in MIPS-ratings (14 for DECStation, 12.5 SPARCstation 1)? In my opinion, no. It gives you a good idea of its capabilities, but if you really want to compare machines, you need to run your code on the various machines to see which is better for your applications. > - how about the software availability, especially AI software? don't know. >If you're going to answer, please do not flame! Keep technical! As far as comparisons between the DS3100 and the Sun 4/60 and Sun 4/330, it really depends on your application. I recently ran several benchmarks on all three and the DS3100 fell, for the most part, between the 4/60 and the 4/330 (closer to the 330). The two areas that the Suns are better were disk accesses and graphics. The disks on the 330 were three times faster than the ones on the DS3100 (I don't know how using a 3rd party disk, like the Wren Runner, would affect this). We ran some ffts that read the data in from a file and wrote the results out to a file. On a Sun 4/330 it took ~10 seconds (real time) to complete the operation. On the DS3100, it took ~26 seconds to do the same thing. When we modified the program to write to /dev/null and read in from a pipe, the Sun 4/330 still took ~10 seconds. The DS3100, though, improved A LOT and only took around 8 seconds. Graphics wise, Sun blows the doors off of a DS3100 because the DS3100 does everything in software wereas the Sun can, assuming you purchase the graphics accelerator, do most everything in hardware. Suggestion: If all you are doing is number crunching, buy the DS3100 (or wait for the rumored DS that will use the newer R3000 MIPS processor that is "supposed" to be out late summer). If you are going to be doing lots of graphics, or lots of things with the disk, I would buy the Sun 4/330 (the 4/60 is nice, but you are probably better getting the 4/330). Disclaimer: The above is my opinion only. Any comments about rumors are just that, rumors and are not to be taken as containing any semblance of reallity. --- W. Tait Cyrus (505) 277-0806 e-mail: cyrus@pprg.unm.edu University of New Mexico Dept. of Electrical and Computer Engineering Parallel Processing Research Group Albuquerque, New Mexico 87131
cquenel@polyslo.CalPoly.EDU (-3 more school days) (06/11/89)
In 10108 gengenba@unipas.fmi.uni-passau.de (Michael Gengenbach) sez: >Now there are two favorized Workstations: > > - DECStation 3100 > - SPARCstation 1 or SPARCstation 330 > >As we do not have much knowledge in RISC-technology, we have a couple >of questions: > > - what do you think about RISC vs. CISC in performance of > AI-applications (Lisp, Prolog, Databases, image-processing), > i.e. symbolic computation? Fer GOSH sakes! The best advice anyone could give: Bundle up some of the stuff you run and tote it on over to whomever you're going to buy from! All you gotta say is : "I'll buy whichever performs better on MY code, not some silly foo-stones!" And let the silly salesmen duke it out. Just be sure you're watching over their shoulder. There could be many reasons why this is impossible, but if it is at all possible, DO IT! --chris -- @---@ ----------------------------------------------------------------- @---@ \. ./ | Chris (The Lab Rat) Quenelle cquenel@polyslo.calpoly.edu | \. ./ \ / | You can keep my things, they've come to take me home -- PG | \ / ==o== ----------------------------------------------------------------- ==o==
cquenel@polyslo.CalPoly.EDU (-3 more school days) (06/11/89)
In 10118 cquenel@polyslo.CalPoly.EDU (0 more school days) sez:
> There could be many reasons why this is impossible,
^ could be
OOOOPS
--chris
--
@---@ ----------------------------------------------------------------- @---@
\. ./ | Chris (The Lab Rat) Quenelle cquenel@polyslo.calpoly.edu | \. ./
\ / | You can keep my things, they've come to take me home -- PG | \ /
==o== ----------------------------------------------------------------- ==o==
aarons@syma.sussex.ac.uk (Aaron Sloman) (06/18/89)
gengenba@unipas.fmi.uni-passau.de (Michael Gengenbach) writes: > Hi folks, > > I know, that this question could mean war, but we really need some > decision support. [.......] > - DECStation 3100 > - SPARCstation 1 or SPARCstation 330 [........] > - what do you think about RISC vs. CISC in performance of > AI-applications (Lisp, Prolog, Databases, image-processing), > i.e. symbolic computation? Our (informal) comparisons between Sun3/280 and Sun4/260 running Poplog (see below) show that on balance the latter is between 1.5 and 2 times faster. On simple Poplog Prolog tests the Sun4 is fairly consistently around twice as fast. But integer multiplication (e.g. factorial(1000)) is slightly faster on Sun3/280! (On Sun386i it was faster still, though other things were slower.) I have not yet tried the M68030-based Suns. My impression is that overall, at present, RISC gives you a better price/performance ratio. > - can one trust in MIPS-ratings (14 for DECStation, 12.5 SPARCstation 1)? The ratios we found between Sun3/280 and Sun4/260 showed that MIPS ratings are at best a very rough guide, and depend on your program. [.....] > - how about the software availability, especially AI software? Our Poplog system (providing Prolog, Common Lisp, Pop-11, ML) is currently available on SPARC machines. It is possible that it will be available on DECstation, but not yet certain, and definitely not before the end of this year. > > Michael Aaron Sloman, School of Cognitive and Computing Sciences, Univ of Sussex, Brighton, BN1 9QN, England INTERNET: aarons%uk.ac.sussex.cogs@nsfnet-relay.ac.uk aarons%uk.ac.sussex.cogs%nsfnet-relay.ac.uk@relay.cs.net JANET aarons@cogs.sussex.ac.uk BITNET: aarons%uk.ac.sussex.cogs@uk.ac or aarons%uk.ac.sussex.cogs%ukacrl.bitnet@cunyvm.cuny.edu UUCP: ...mcvax!ukc!cogs!aarons or aarons@cogs.uucp
hurf@batcomputer.tn.cornell.edu (Hurf Sheldon) (06/19/89)
Steve Thompson, olin@cheme.tn.cornell.edu, has done some fortran comparisons of the 3100 and the Suns and his results indicate that in most instances the 3100 is somewhat more efficient than the Sun Sparc. Please contact him for details. hurf -- Hurf Sheldon Network: hurf@ionvax.tn.cornell.edu Lab of Plasma Studies Bitnet: hurf@CRNLION 369 Upson Hall, Cornell University, Ithaca, N.Y. 14853 ph:607 255 7267 I sold my Elan, got a job in science; Now, no one takes me seriously.
avr@cs.purdue.EDU (Andrew V. Royappa) (06/21/89)
In article <1094@syma.sussex.ac.uk>, aarons@syma.sussex.ac.uk (Aaron Sloman) writes: > Our (informal) comparisons between Sun3/280 and Sun4/260 running > Poplog (see below) show that on balance the latter is between 1.5 > and 2 times faster. On simple Poplog Prolog tests the Sun4 is fairly > consistently around twice as fast. But integer multiplication (e.g. > factorial(1000)) is slightly faster on Sun3/280! (On Sun386i it was > faster still, though other things were slower.) > I noticed that multiply and divide on my Sun 4/110 were not as fast as I expected them to be. Disassembling the code showed that integer multiplication and division were actually not done in hardware, but as subroutine calls to library routines. Just out of curiosity, why doesn't the Sun 4/110 not have hardware integer multiply and divide ? Is this part of the "RISC" concept ? Thanks, Andrew Royappa Purdue University {ihnp4,pur-ee}!purdue!avr avr@purdue.edu