Xref: utzoo comp.sys.dec:941 comp.arch:7896 Path: utzoo!attcan!uunet!seismo!sundc!pitstop!texsun!jthomp From: jthomp@texsun.Central.Sun.COM (Jim Thompson Sun Dallas SWAN Engineer) Newsgroups: comp.sys.dec,comp.arch Subject: Re: DECstation 3100 info. Message-ID: <350@texsun.Central.Sun.COM> Date: 17 Jan 89 21:17:15 GMT References: <979@isieng.UUCP> <85330@sun.uucp> <1798@ardent.UUCP> <85553@sun.uucp> Reply-To: jthomp@hemaneh.Central.Sun.COM (Jim Thompson Sun Dallas SWAN Engineer) Organization: Sun Microsystems Inc., - Dallas, TX Lines: 46 In article <85553@sun.uucp> khb@sun.UUCP (Keith Bierman - Sun Tactical Engineering) writes: >In article <1798@ardent.UUCP> mac@mrk.ardent.com (Mike McNamara) writes: >>| [discussion of Mips chips in various boxes.] >> >> And it makes sense, The Ardent Titan is up to four Mips R2000 >>each tightly intergrated with a custom vector processing unit. What do >>you expect an M120 to do when given an Ardent `dvma' instruction? >>(double precision A = B*C+D). (Probably bus error...). >> > >C'mon old buddy, isn't there a way to generate scalar only code ? >Shouldn't one expect scalar only code to be portable ? A vector >version of gnu-emacs, for example, might be nice...but being able to >"plug and play" can be handy. Turns out that a vectorized gnu-emacs is a loose. The only place you pick up speed is in bcopy, and then only for strings longer than the vector length. (At least that's what I found at Convex... :-) Overall, the vector version of gnu emacs was slower than the version compiled without vectorization. (And the version compiled with gcc was better than either!) I have typically found that vectors and vector machines are usually left to the pipe stress freaks and other fp grunts. But then, its been a long time since I wanted to write any Fortran. :-) In most cases, the context switch overhead of getting on the vector board(s) isn't worth the speed increase unless you can keep the dern things fed for some large (relatively speaking) amount of time. For general purpose computing, gimme something like what the people at Prisma are trying to do. (fast GaAs, Unix, SPARC.) (Did that sound like an ad? Sorry.. :-) Disclaimer: I work for Sun. I used to work for Convex. I have friends who work for Prisma. I don't own stock in any of the mentioned companies. I don't speak for Sun, Convex, or Prisma. This has been a recording. -- Jim Thompson jthomp@central.sun.com "I woudn't recommend sex, drugs, or insanity Network Engineering for everyone, but they've always worked for me." Sun Microsystems -- Hunter S. Thompson