Re: Poor Sun timings, as competition...

[GRoberts@YUKON.SCRC.Symbolics.COM (Gary Roberts)]

    Date: Mon, 6 Aug 90 18:22 EDT
    From: GRoberts@YUKON.SCRC.Symbolics.COM (Gary Roberts)

	Date: Mon, 6 Aug 90 15:41 EDT
	From: Jeffrey Mark Siskind <Qobi@ZERMATT.LCS.MIT.EDU>

	    Date: Mon, 6 Aug 90 13:28 EDT
	    From: Gary Roberts <GRoberts@YUKON.SCRC.Symbolics.COM>

		Date: Tue, 31 Jul 90 18:19 EDT
		From: Qobi@ZERMATT.LCS.MIT.EDU (Jeffrey Mark Siskind)

		    Bubba sez: M-sh-C.  And as he has said before, ignorance is truly bliss
		    (even the truth is relative). 

		Most of the time, recompiling an entire 600 line file under the Lucid
		development compiler on a Sparcstation I+ will be faster then recompiling
		a single function on a 3640 with m-sh-C.
			Jeff

	    That's pretty interesting.  Is there someone out there who could tell me
	    how fast an XL1200 is vs. a Sun 2? :)

	Pretty funny. You are missing my point however. It is very unlikely that many people
	will buy XL1200s. As much as I would like one, MIT will not buy a
    >$45K machine (XL1200)

    I have been surprised by the success of the XL1200 so far.  Last month,
    the first month of availability, we sold over half as many XL1200s as we
    sold XL400s over that product's entire lifetime! Even your AI Lab bought
    a couple.  I expect we will sell more there as time goes by.

	that performs about the same as an <$10K machine (SparcStation I+). I will strengthen
	my claim:
	  Most of the time, recompiling an entire 600 line file under the Lucid
	  development compiler on a Sparcstation I+ will be faster then recompiling
	  a single function on an XL1200 with m-sh-C!

			     
     I apologize for my joke.  I agree that the Symbolics compiler is
    probably slower than the Lucid compiler, but your statement above sounds
    suspiciously like a "flame".  Have you had the opportunity to get much
    time on MIT's non-existent XL1200's in order to extensively test your
    assertion?

	    Yes, it is true that the development environment on Symbolics is better than on a Sun
	but the gap is closing. While Genera 8.0 is an improvement over System 98.0, everybody
	else is also improving, and at a rate faster than Symbolics. Though they are not yet
	there, they are quickly catching up. In 1982 the claim that the Symbolics development
	environment was light years ahead of the competition was truly valid. It isn't so
	any more. What the m-sh-C example shows is that sometimes raw speed more than makes
	up for lack of a fancy feature. IMHO there is a lot of latent speed left in Symbolics
	machines, both the XL1200 and even the lowly 3640---if only a decent compiler was
	written for it. Two months ago I would never have thought of giving up my Symbolics
	machine. I was a staunch defender of it. But two months of hacking Lucid on SparcStation I+
	has made me think twice. Speed may not be the only thing but it sure is an important
	factor. Price/Performance is another. You can't live on your laurels forever.
		Jeff

    The XL1200 has been successfully benchmarked (by both Symbolics and
    users) against the SPARC/1, the 4/330 (a $40K+ system), and 4/370 (a
    $50K+ system), and I am happy to report that we handily beat all of
    those systems in most cases, including execution of the
    "not-a-benchmark" that you yourself published a few months ago, and
    which were run recently by Jeff Siskind of Xerox PARC. While it is true
    that Lucid (but not Franz) on these systems sometimes compiles faster
    using their development setting, it is not nearly as much as you
    suggest, and the XL1200 makes up for it with a substantial advantage in
    execution speed. 

Editor's Note:  I realized while driving home last night that I had
confused Jeff (the "not-a-benchmark"'s author) with Mark Shirley of Xerox
PARC, who ran the tests on the XL1200.  That was as bad a mistake as
mixing up a 3640 and an XL1200!  Now, if it turns out that Mark is three
years older than Jeff and runs six times slower, my face will really be
red!!!