home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mitech!gjc
- From: gjc@mitech.com (George J. Carrette)
- Newsgroups: comp.lang.scheme
- Subject: Re: Are interpreters now as fast as compiled code used to be?
- Message-ID: <4078@mitech.com>
- Date: 16 Dec 92 17:23:31 GMT
- References: <4051@mitech.com> <FEELEY.92Dec14215701@zohar.ai.mit.edu> <4067@mitech.com> <FEELEY.92Dec15153031@zohar.ai.mit.edu>
- Organization: Mitech Corporation, Concord MA
- Lines: 48
-
- In article <FEELEY.92Dec15153031@zohar.ai.mit.edu>, feeley@zurich.ai.mit.edu (Marc Feeley) writes:
- > Well a factor of 3 for the last decade is clearly not valid but for
- > the next 5 years it might be. DEC's Alpha introduced this year (200 MIPS
- > or so) is about 3 times faster than last year's HP9000/700 (70 MIPS).
- > If next year DEC comes out with their BIPS processor (1000 MIPS) my
- > factor of 3 seems about right.
-
- You need to track a given manufacturer, not jump from one to another,
- because the computer guys are not -really- leapfrogging each other.
- They just appear to be for marketting purposes. Also "benchmark rot"
- is involved as people play games with funky preprocessors that don't
- always mean much for what good programmers would write.
-
- [Also, people with real work to do hardly ever win by jumping from
- one machine manufacturer to the next ever year or two].
-
- Look at what SUN has done with 680xx and SPARC since say 1983 to present.
-
- And HP the same, starting with their 680xx machines and their new RISC line.
-
- And for the longest time curve, you have the DIGITAL VAX line. Where this
- years VAXSTATION is about the same speed as last years HP RISC machine.
- And this years lowest-end ALPHA about 3 times faster than that. Ah,
- the famous factor of 3.
-
- > Clearly this increase in performance will level off at some point in
- > the future.
-
- No, I think we have seen a pretty steady factor of 100 over the last
- decade. That works out to be about a factor of 1.6 per year.
-
- Glitches are caused by project start up, manufacturing, and other
- lead-time delays. Where you can get shifts of two years, typically.
- And 1.6 squared is about your factor of 3.
-
- > When that happens the additional factor of 100 or so
- > provided by compilers will be more valuable.
-
- I don't agree that compilers give a factor of 100.
- On simple numerical stuff, yes. But that is easy operator
- specialization. Even interpreters can do that.
-
- > Note that parallel
- > processing does not change the argument because what counts is the
- > number of MIPS you can get with your budget.
-
- Bring back timesharing?
-
-