tackett@ipla01.hac.com (Walter Tackett) (03/30/91)
From: tackett@ipla01.hac.com (Walter Tackett) Newsgroups: comp.ibm.pc.hardware Subject: The REAL speed of a Weitek. Summary: Followup-To: Distribution: na Organization: Hughes Aircraft Company - Canoga Park, CA Imaging Guidance Design Laboratory Keywords: Thanks for collective responses to my EISA/ISA question (i will summ- arize the voluminous replies by mail to anyone interested). Next question: What is the actual speedup factor provided by a Weitek 4167 @33Mhz relative to the 486 it is slaved to *GIVEN* that I am running tight do-loops of mutiply-accumulate operations on contiguous memory elements with no branches or IO ops (eg convolutions, DCTs, matrix multiplies), assuming that i have a c or (*ick*) fortran compiler which supports the 4167. There was an article in PC mag which gave some useful information, but wound up making me ask more questions: the 4167 *they* tested didn't provide any speedup because they used it with an application which did alot of I/O that somehow floundered the Weitek. They also made a claim that a 3167 actually has an advantage over the 80387 since the former is memory mapped (faster xfer) instead of IO mapped. I'm pretty sure that IO mapping will not be advantageous over the "80487" (sic) internal to the 80486. I have heard people quote the chip specs as being 10-15 times faster, but i'm willing to bet that that's an unattainable theoretical upper limit. SO if you have any advice, references to people, articles, USENET topics, etc, let me know since i would like to get some kind of idea what to expect before ask- ing these questions to people who have an interest in selling me chips and compilers. $1000 is a terrible thing to waste :-( Thanks in advance, -Walter