Path: utzoo!mnetor!uunet!husc6!bloom-beacon!gatech!mcnc!ecsvax!hes From: hes@ecsvax.UUCP (Henry Schaffer) Newsgroups: comp.arch Subject: Re: Performance increase - a suggestion Message-ID: <4554@ecsvax.UUCP> Date: 3 Feb 88 20:08:24 GMT References: <235@unicom.UUCP> <28200089@ccvaxa> <3127@phri.UUCP> <230@m2.mfci.UUCP> Organization: NC State Univ. Lines: 27 Summary: too true, accuracy of results doesn't sell In article <230@m2.mfci.UUCP>, root@mfci.UUCP (SuperUser) writes: > ... > My personal perception of the market for scientific computation is > that given a choice between more precision and more speed, speed > wins hands down. > > Bob Colwell > Multiflow Computer > 175 N. Main St. > Branford CT 06405 mfci!colwell@uunet.uucp 203-488-6090 I agree, but suggest that this is a result of the triumph of marketing over good sense. More speed *always* is useful, and there is a limit to how much precision one needs, so there clearly is a domain in which more speed is more useful than more precision - but I don't believe that the usual single precision (4 byte) floating point is in that domain. If I'm right, then one should design for speed rather than precision - shouldn't one? (One could argue that the smaller word decreases storage needs as well as increases speed - which is correct, but one can find many examples of choice of algorithms in which the tradeoff between speed and accuracy doesn't significantly affect storage requirements, and yet people either choose speed or don't take the effort to choose accuracy.) --henry schaffer n c state univ