Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Brilliant....#4

Author: Robert Hyatt

Date: 05:55:16 05/09/04

Go up one level in this thread


On May 09, 2004 at 01:15:06, Dave Gomboc wrote:

>On May 08, 2004 at 22:01:47, Vincent Diepeveen wrote:
>
>>On May 08, 2004 at 21:37:00, Robert Hyatt wrote:
>>
>>>On May 08, 2004 at 21:14:04, Vincent Diepeveen wrote:
>>>
>>>>On May 08, 2004 at 21:04:19, Robert Hyatt wrote:
>>>>
>>>>You are just flaming as usual.
>>>>
>>>>Show us the outputs of cray blitz. No one beliefs you. Your thesis is unfindable
>>>>too. Show it.
>>>
>>>It is available through University Microfilm.  I've told you that previously.  I
>>
>>Where must i email to to get a copy of that? Going to alabama myself is a bit
>>far. paying for reproduction costs is no problem.
>>
>>I have like 10 programmers here in europe who all are looking forward to seeing
>>it.
>
>Now you're embarrasing yourself, Vincent.  Really.  This took about 3 seconds to
>find with a Google search.
>
>http://www.umi.com/hp/Support/DServices/order/
>
>Dave


Haven't you realized yet that he isn't really interested in getting a copy of my
dissertation, which is easy enough.  He is only interested in producing yet more
of his patented brand of disinformation.  By claiming that it is unavailable,
whether it is or is not is unimportant.  I have used this service several times
in the past.  I have explained how to use it several times.  He doesn't really
want a copy, because he knows that the DTS article provides the same basic
description.  And he knows that my dissertation uses the BK positions which he
claims are not good tests, being too easy to get a 4.0 speedup on with 4 cpus.
I found the BK positions were a bit harder in the data logs I posted here last
week.  But real data won't deter him.

He claimed I wrote an article for JICCA that not one single person can find.
He claimed that I wrote in CCC posts and in that JICCA article that my speedup
approximation formula worked for any number of processors, where Rolf posted one
CCC post that directly contradicted him.

He claimed that my speedup formula was wrong, not understanding the concept of
"linear fit to a non-linear set of data.  A bunch of Opteron test data showed
that the formula was also reasonably close and also that it even underestimates
the speedup at times due to variability.

Once he saw the data, he switches to claiming that the 1cpu version could be
faster by 10%, making the data look better than it should.  Where 10% came from,
I can only guess.  I would guess the exit of his digestive tract myself, since
when I first changed Crafty for the SMP stuff I measured that and posted the
numbers here.

I even modified my data to account for his claim, even though I know that the
program was not slowed 5%, particularly on the opteron with 8 extra registers.
He neglected to think about this "claim" as a normal person would do and realize
that 5% will turn into 5%/3 slowdown with 3 processors.  But he neglects to
think about most of what he writes.

But real data doesn't matter.  Only disinformation.  He'll never cite the JICCA
article because it doesn't exist.  He'll never provide the exact quote by me
saying that the speedup formula works for any number of processors because it
doesn't exist.  He'll never admit that my speedup approximation is reasonably
accurate for 1-8 (or even 1-4) processors because it isn't in his best interest
to do so even though a ton of data shows that it is so.

There is a trend there...

Just like the claims that his program has the best evaluation.  The best search
The best search.  That Hsu's singular extension idea is no good.  Only to later
claim to have it implemented in his current code.

disinformation.  _Intentional_ disinformation.

Browse r.g.c.c to find his 1995 nonsense about "Diep is the best correspondence
engine in the world" and the like.

10 years worth of disinformation.  He dislikes me because I have this horrible
habit of simply running the test and posting the results..

He claimed that null-move was the reason he couldn't get reasonable speedup
numbers.  That null-move affects the speedup "bigtime".  I said "no" after
running a pretty quick test.  He said "you didn't even run a test, you just make
up an answer and post it."  I immediately supplied the sample data.  He responds
"you only used a position or four.."  I asked "what test would you like?"  He
said "use the DTS article positions, crafty produces no speedup on my dual using
those."  I ran them on my quad and got 3.0X faster.  Null move disabled produced
a 3.1 speedup.  GCP ran the same test with null and got 2.8X.  He writes "GCP
proved that your formula is dead wrong."  Also that 3.0 vs 3.1 proved that
null-move makes parallel search less efficient bigtime.  I just didn't realize
that .1 was "bigtime" I suppose...

This will probably never end...  But at least I plan on continuing to provide
real data to counter all the disinformation.

Big problem is that Vincent doesn't realize how badly he keeps embarassing
himself...  Particularly in light of differences between "real performance"
versus "claims" as it applies to his wild "claims".



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.