Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: WMCCC - may the best man at getting the fastest hardware win :(

Author: Robert Hyatt

Date: 10:12:48 10/19/97

Go up one level in this thread


On October 19, 1997 at 10:54:34, Chris Whittington wrote:


>
>P11/300's might be 300/233 better than K233's, but memory bandwidth
>indicates less. Certainly less than the 28% implied. Maybe 15% or so.
>Yoru 750 Mhz alpha is reported to give a 3.5 multiplier to your node
>rate.
>
>This is a large difference and is the point you allegedly missed above.
>

Ok... then time for the next question:  "how fast is too fast?"  IE
why is it ok for Fritz to run on a PII/300, yet we should use a K6/233?
I don't buy that.  And btw I have *no* idea about the multiplier for the
alpha/766...  the processor is cooled way down, and probably run at an
elevated Vcc to match the elevated clock speed.  But no one knows how to
make memory any faster as of yet, so it might not be any faster at all
for all I know.  I have a very low probability of running on one of
these
anyway.  DEC is giving us a "vanilla" alpha/500mhz machine.  Not a
super-
alpha (600mha) nor anything special... just the sort of machine you
might
buy from Polywell for under $5,000...


>Additionally, you could perfectly well run (as could Bruce and as could
>Fritz) on the K233's provided.
>
>So the point is that YOU and Bruce (and less so Fritz) are taking the
>decision to try and massively alter the result of this event. You're
>doing it under a screen of everybody else either does it or has done it
>and nobody complained in the past and its all quite normal and and and
>and infinitum; and in the past it was those terrible COMMERCIALS and
>nobody complained, but now we are AMATEURS (with resources of course),
>so its ok, and anyway AMATEURS are better than  COMMERCIALS both in
>terms of the programs and morality.
>
>So why don't you get together with Wullenweber and all agree to use the
>K6's ?


I think it is too late.  Dark Thought is an example.  I don't know
whether
they can run on a K6.  I doubt they will.  Fritz won't run on the K6.
All
I'd say is that if *everyone* ran on K6's, I'd be more than willing to
run
on a K6.  But if *anyone* runs on an alpha, I want an alpha.  If
*anyone*
runs on a PII/300, I want a PII/300.  I don't want an advantage over
*anybody*...  but I don't want to be at a computing-power disadvantage
either...




>
>Then the rest of us can participate and observe a rather more fair and
>meaningful tournament where nobody has bought or maneouvred themselves a
>massive advantage over the rest of the field .........
>
>>
>>
>>>b) you never suffered competing against them since neither Bob nor Bruce
>>>have ever played in a WMCCC where these manufacturers machines appeared
>>
>>
>>I've run into this at ACM events.  Same idea.  CDC whipped everyone in
>>the
>>early 70's with the Cyber machines that were 10x faster than the next
>>fastest
>>machine in the tournaments.  We just kept working and trying to find a
>>machine to even things up...  we didn't complain...
>
>You couldn't complain. The ACM is totally open to mainframes, parallels
>and whatever. Apart from the odd micro (which often created nice upsets)
>this ACM was a playground for academic empire builders with too much
>time and too much computing power on their hands. Its irrelevant to the
>current discussion, the games were not of good quality, it was a closed
>club. Quit trying to pretend that parallels can be drawn to the WMCCC.

hmmm... poor argument.  The WMCCC is totally open to "any commercially
available microprocessor."  So the two events now have their parallels.
I didn't like the "arms race" of the ACM and WCCC events.  But I wanted
to
participate and have a chance, so I joined in.  Again, get everyone
together
an Paris and amend the rules for next year:  entrants *must* use the
supplied machines.  Doesn't bother me for everyone to have equal
hardware.
It's not completely fair, still, because it nullifies roughly 3 years of
work on my part.  But the *next* one (1999) will almost certainly be
played
on 64 bit processors anyway, so I get my optimizing work back...



>
>>
>>>
>>>c) If you don't approve of the idea of these manufacturers getting an
>>>edge over a basically 'uniform' rest of field (as was the case in those
>>>days), then why are you doing exactly the thing you list you disapprove
>>>of then, now ? Because, without these very fast alphas, this Paris WMCCC
>>>would be, to all intents and purposes, a level playing field.
>>
>>
>>not from where I sit.  A PII/300 is not "roughly equal" to a K6/233.
>
>Please don't try and fool us that your apparent 3.5 times is remotely
>comparable to the 15-30% or so of a PII/300

and please don't try to fool me that 30% is "trivial".  I run such tests
all the time, and can absolutely measure the difference in performance.
It
is not something I'd give away for nothing.  IE a loop stuck in my
move generator to slow things down by "only" 30%.  It would be a
significant
advantage...  And again, how fast is too fast?  Either go *all the way*
and
let's do it uniform, or else choose to play or not play as things exist
now.  Because I'm sure if Bruce and I were on PII/300's, this topic
would
still be going on...

and you haven't touched on the Dark Thought/Chess Guru problem either.
What about them?  I commented before last year's event, and the year
before
that, that I didn't think alphas belonged.  That was pooh-poohed by
several
programmers. So now I am on one of them too.  This is not a "career" for
me, this is fun.  I'm not going to discuss and beg for machine time, and
so forth.  I'm simply going to follow the rules as they are written, and
do what has been done since the very first WMCCC event was held over a
decade ago.  I don't mind changing the rules, but for next year.  DEC is
paying *everything* to ship us an Alpha to Paris.  They have been very
helpful, hoping for some positive publicity out of this.  One week
before
the event is too late to change things.  At least two more would
probably
have to withdraw, and there would be a hue and cry from those that are
bring PII/300's...



>
>> A
>>hand-coded assembly engine optimized for this is an even bigger edge.
>
>We've discussed this over and over. Assembler is faster, nominally, but
>
>a) its buggy
>b) its hard to maintain
>c) compilers often do a better job than hand-coders now
>d) its takes longer to code
>
>Its an open question whether assembler helps. I can recollect Bruce
>commenting that C was better for coding, better for debugging and that's
>why he used it. The portability issue is a more recent one that he has
>brought up, but wasn't his stated reason last year.

It's not an open question with me.  Cray Blitz (pure fortran) is 5X
slower
than Cray Blitz + 20,000 lines of hand-coded/optimized assembly code.
5X
is a hell of a bonus.  What Fritz gets from asm is unknown...  but I
doubt
it is a few percent.  Of course, I can get 300% by switching platforms,
because I made a conscious effort to remain portable for this very
reason.


>
>> I
>>wrote crafty specifically to (a) be portable; (b) to take advantage of
>>64
>>bit machines.  So it's ok for Fritz to hand-code and tune and show up
>>with
>>a faster processor?  I'd bet I am no faster than him if I were to run on
>>the 766 mhz machine.  Maybe not even as fast...
>>
>
>Fritz is a root processor with practically zero evaluation function.
>therefore he is very fast.
>
>Crafty is a tip node processor with slightly more in the eval function.
>Therefore, on equal platforms, Fritz will be faster in nps terms.
>
>If you're trying to pretend that your 750 Mhz alpha just puts you on an
>equal footing with Fritz nps-wise, then your argument is shallow.

I'm not sure it does or doesn't.  I don't know.  I do know I search
around
200K on the 500 mhz alpha we will probably be using.  I don't know if
the
766 will be any faster at all.  200K is probably significantly slower
than
Fritz on the PII/300...  There are only *two* Kryotech machines promised
for this event, Dark Thought and Bruce.  Had we asked before Bruce
thought
of doing so, we would be on one and he would be using his 533.  Chances
are
> 90% that we will use a 500mhz alpha.  A processor that has been out about
as long as the P6/200.  It just happens to like Crafty, because it is a
true
64 bit app.



>
>You're not trying to be on an equal footing with Fritz or anybodty else,
>you're trying to get a massive hardware advantage, quoted at 3.5 times.

You look at it as though I want to get way ahead of everyone else.  That
is *wrong*.  I'll play in any uniform-platform event you care to
organize.

What I am *really* doing is trying to negate a significant advantage
that
others would have over *me* were I on a K6.  At least 1/3 and maybe as
much
as 1/2 of the field is going to be on faster machines than the K6.  I'm
trying to eliminate that edge, not gain one of my own.



>
>>
>>>
>>>That it is not a level playing field now is pretty much a Bob/Bruce
>>>issue. I guess we can live with the odd K6 at 266 or 300. 300 to 233 is
>>>not the advantage the raw numbers imply (because of memory bandwidth). I
>>>can live with a tournament where 'effective' speeds vary by 20, 30 % or
>>>so. Its when you can get 2x or 3x by running oin these alphas that I cry
>>>foul.
>>
>>
>>your counting is way off.  Multiple PII/300's, *four* programs running
>>on
>>an alpha.  Maybe more.  Two years ago there were Sparcs entered that
>>were
>>faster than the Intel machines of the time.  It is *not* just Bruce and
>>I.
>>I got interested in faster hardware when it became apparent that we had
>>to
>>to remain competitive...
>>
>
>This is unbelievable use of english. You're not 'trying to remain
>competitive'; you're' trying to get a bloody great advantage' (3.5 x).

and if Bruce/I didn't do this, we would be 3x slower than Dark Thought.
1.5X slower than Fritz and others on the PII/300.  I'm not sure why we
are suddenly the "bad boys" here, we got into the alpha market way late.
Check out who ran on 'em last year.  And in the French Championships the
last two years...  We "joined" the party, we didn't create it...



>
>>
>>
>>>
>>>>Each of these
>>>>architectures offers different features.  Now on to the alpha...
>>>>
>>>>If you go back to the beginning of my work on Crafty, which dates back
>>>>almost
>>>>exactly 3 years ago, now, you know that one of my specific goals for
>>>>this
>>>>"experiment" was to target development for a 64bit architecture, such as
>>>>the
>>>>Cray *or* the alpha.  Most of you have written your programs optimized
>>>>for
>>>>16/32 bit words.  I wrote mine optimized for 64 bit words.  It doesn't
>>>>seem
>>>>reasonable to me that Crafty should have to be handicapped by having to
>>>>run
>>>>on a non-optimal architecture, "just because everyone else is."  So that
>>>>is
>>>>point 1:  a program ought to run on the architecture it is designed for.
>>>
>>>So why not run it on a Cray then ?
>>>
>>>Because that's outside the rules.
>>>You're just going outside the spirit in order to get an advantage. And
>>>no mean advantage either. Look, Bob, I want to see, and I suspect others
>>>want to see, a tournament where the software gets judged; not one where
>>>some people have manoeuvred or bought themselves a massive hardware
>>>advantage.
>>>
>>
>>
>>I'm game.  Get the programmers together in Paris.  See if everyone wants
>>this to be a uniform platform event.  I don't mind either way.  If most
>>want a uniform platform event, then approach the ICCA and recommend it.
>>The ones that don't like it can play somewhere else.  But I'd want it to
>>be uniform for *everyone*, and not *most*.  IE Fritz runs on what we run
>>on... so does genius, rebel, hiarcs, CSTal, etc...
>>
>>
>>>
>>>> I
>>>>don't know of any way to "equalize" the alpha and the X86...  it depends
>>>>on
>>>>the program.  Bruce gets 1.8x faster on his 533 than on the P6/200,
>>>>Crafty
>>>>seems to get exactly 2.5X faster on a 500mhz machine than on the P6
>>>>based
>>>>on recent tests.  That's because I'm designed to pump 64bit words
>>>>around,
>>>>while Bruce is not.
>>>
>>>So you guys want to win at all costs. Makes much of the optimisation and
>>>work of all the other guys running on K6's seem a little futile, doesn't
>>>it ?
>>
>>depends on where you look.  What about the 3 years *I* have spent
>>writing
>>a program that takes advantage of 64 bit architectures?  So I throw all
>>*my*
>>optimizing out, so that your optimizations give you an edge because of
>>the
>>architecture you have optimized for?
>
>I don't have any 'optimisations'. I write CSTal from the point of view
>of a chess player, not a programmer. My time gets spent studying games,
>not looking at bits and bytes.
>
>> And that is somehow supposed to
>>make
>>me feel that this is somehow "fair"???
>
>Its 'fair' right now that we all run on roughly equivalent systems. Then
>we, and joe public, can read something into the results. That's fair.
>
>Suppose you or Bruce win. What will that mean ? That you have the best
>program ? No, not necessarily. Not necessarily at all.



and what if we lose?  then we must really suck.  It is a risk no matter
how we look at it.  I don't like the arms race.  I didn't start the arms
race.  But, as the saying goes, I want to bring as big a stick to the
fight
as anyone else brings...

>
>>
>>
>>
>>>
>>>>
>>>>
>>>>Second, alphas are *not* expensive.  They can be had for under $5000,
>>>>which
>>>>is less than I paid for my P6/200 18 months ago when I got it.  So this
>>>>is
>>>>*not* exotic hardware.  point 2.  It is readily available...
>>>
>>>Come on. Its a universtity research machine. The market for selling
>>>chess programs onto it is precisely zero. Its just a way to get a
>>>massive advantage by spending several thousand dollars, and,
>>>incedentally claiming to be an 'amateur' programmer. 'Amateur'
>>>programmers don't work like this, IMO.
>>
>>you should talk to Polywell, Enorex, and multiple other vendors.  they
>>are selling all they can build.  I know of several at a couple of
>>businesses
>>here in Birmingham.  Since they run NT, and most microsoft products are
>>being ported, it is a viable platform that offers a higher level of
>>performance
>>than the intel platform offers...
>
>Great. Fine idea. Wish them all the best.
>
>But why don't you put them away and run on the other system that your
>programs run on (namely PC), just for this tournament, just for fairness
>?

get everyone to.  I'll sign up immediately.  But *only* if *everyone*
does..



>
>>
>>
>>
>>>
>>>>
>>>>Next, the overclocking issue.  It was OK for Lang to overclock Mephisto,
>>>>it
>>>>was ok for Fedility to overclock chess challenger, and for Dave to use a
>>>>bit-slice (think it was him but I'm not sure) for what was back then
>>>>called
>>>>various things such as "superconny"...  So I don't buy the fact that
>>>>suddenly
>>>>Bruce or Ernst have done something unconscionable here.  Crafty *might*
>>>>get a
>>>>766 machine.  It will definitely get a 500mhz machine.
>>>
>>>You and Bruce were never affected by overclocking Mephistos or anything
>>>else, because you WEREN'T at those tournaments. This is a criminal
>>>argument to bad propose actions now on the basis of bad actions in the
>>>past.
>>
>>So we watch what goes on for 10 years, with nary a complaint from
>>anyone,
>
>You're joking ?
>
>There were 750 Gigahertz of complaints. But, as ever, the powers that be
>didn't pay any attention, in fact they encouraged it with manufacturers
>prizes and this prize and that prize and the other prize.
>
>However the programmers and critics and so on were all dead against it,
>and all said so, often.
>
>>but when *we* follow the same path, *WHAM* we get slapped for it?
>
>I believe this is ad hominem. This has nothing to do with who you are,
>more to do with the fact that our friendly internet forum allows us to
>discuss. You just happen to be the transgressors right now. Doesn't mean
>the old transgressors were condoned at the time.


we are but *two* of close to 16 "transgressors" in this event.  IE 1/2
of the field probably won't use K6's.  So it is not even clear that a
majority will want a "uniform platform" event.  I vote "yes" if my vote
is needed.  But I'll play under the current rules if that is the choice
of the majority...

>
>>
>>Consistency is everything.  If you don't like this, lets change the
>>rules
>>and we can all run on identical hardware.  But if one other program is
>>going
>>to use something different, I am too if I can find something I like...
>>
>
>This sounds like the classical excuse for an arms race.
>
>Arms races come about through lack of trust, lack of communication, and
>desire to get there before the other guy.
>
>OK, then settle it with discussions with Bruce and Wullenweber. I bet
>you could all agree to go on K233's


but there are 13 other programs running on alphas and PII/300's... it is
too late to change now.  The ICCA didn't ask for enough machines for
everyone...

>
>>
>>
>>>
>>>That was then. Now is now.
>>>
>>>Now  is pretty much a level playing field. You two are the guys with
>>>this year's hydraulic jacks rasing the ground by a 45 degree slope away
>>>from your own goal.
>>>
>>
>>again it is not "you two".  there are at least 4 alphas, and 8 PII/300's
>>I know of so far, with a couple of "surprises" that very few know about
>>that will show up on these boxes.  You make it sound like we are the
>>only
>>two, but that's a false impression...
>
>Can you really not live with the odd P11/300 ? Its not a big deal. YOU
>with 750 Mhz alphas are the big deal.
>

I probably won't have one.  I'll probably use an off-the-shelf 500mhz
machine just like bruce uses on ICC roughly...



>>
>>
>>>>
>>>>next the concept of "carrying a custom machine under our arms."  If you
>>>>check
>>>>at the last WMCCC, *many* toted their own machines and it raised no
>>>>eyebrows.
>>>>I think part of this is caused by the "shrinking gap" between commercial
>>>>and
>>>>amateur programs.  But, in crafty's case, we aren't carrying *anything*
>>>>to the
>>>>tournament.  DEC is supplying us a machine.  Kryotech will if possible.
>>>>But
>>>>this sounds *just like* what AMD is doing as well.  I'm not buying an
>>>>alpha
>>>>at present.  I'm using one supplied by DEC.  Others are using PC's
>>>>supplied by
>>>>AMD or whomever.
>>>
>>>Yeah, yeah, yeah.
>>>
>>>Except all your programs (with a few exceptions of knowledge ones) are
>>>just the same, aren't they ? They all use recursive null move and and
>>>and and. They all do a little evaluation at the tips. Some do a bundle
>>>of pre-processing and very little evaluation at the tips. You all know
>>>and use the same tricks. None of you have got any real edge over the
>>>others. Any of these programs could win - it will be down to luck and
>>>machine speed.
>>
>>
>>what's this got to do with anything?  I didn't know there were rules
>>that
>>said "not more than N null-move programs are allowed."
>
>Not saying that. I'm saying that the field contains so many conceptually
>similar programs (either crafty clones or crafty concept clones) that to
>get ahead you need luck or hardware.
>
>Which is why you want 750Mhz.
>

please don't psycho-analyze my motives.  I am getting an alpha *after
the
fact*...  Dark thought plays well.  I didn't want to see them outrunning
us significantly again, as they were last year.  Ditto for PII machines
for
others...  I'm doing this to *stay even* but *not* to "get ahead."



>The knowledge programmers work differently.
>
>>
>>>
>>>So two of you are going into an overkill on machine speed to try and
>>>win.
>>>
>>>Either by spending resources (not exactly amateurish), or by using
>>>contacts (also not exactly amateurish).
>>
>>
>>both of these are wrong.  I spent *nothing*.  I bought myself a P6/200
>>for my home machine that everyone here uses.  I have a P6/200 in my
>>office.
>>I've been using this machine for 1.5 years and paid 5500 bucks for it
>>last
>>year.  The machine in my house I bought for 2000 bucks.  I did *not*
>>make
>>the alpha arrangements.  Jason Deines is gong to Paris to operate
>>Crafty.
>>He and I have been "internet friends" for a year or longer.  He took it
>>upon himself to call Digital Marketing, and pursue a faster machine.  He
>>knew *no one* there.  Neither did I.  So there were no special contacts,
>>just some money invested on phone calls.  You could have done the same
>>thing
>>if you wanted, with the same success we had most likely...
>>
>
>I don't want. I'm quite happy to be on an optimal normal system, thank
>you.
>

so would I be if *everyone* used one.


>>>
>>>Its really as simple as that, isn't it ?
>>>
>>>May the best man win :(
>>>
>>>
>>>
>>>>Somehow I get a sense of "those damned rich amateurs
>>>>vs us
>>>>poor commercial programmers" and I simply don't "get it."  I'm not rich.
>>>
>>>I think I'll wait until after Paris to open up the issue of full-time
>>>paid academic work on computer chess being classed as 'amateur'. You
>>>know as well as I do the numbers of careers in academia dependent on
>>>computer chess programming. The numbers of contracts given out for
>>>research including creation of programs or testing the B***** algorithm;
>>>and the empires built on it.
>>
>>You are taking the wrong angle here.  I am not a "full-time chess
>>person."  I work on Crafty mainly from 9pm to 1am every night.  I
>>do many other things during the day, from teaching, to running a
>>large computer science network/lab, to doing research in various
>>project areas like genetics, parallel programming, data mining and
>>other things.  Computer Chess has *never* been anything but a part-time
>>fun project for me.  Sorry, but I'm not Hans Berliner, nor the DB guys.
>>I wish I could do it full-time, but I don't...  so your point has *zero*
>>merit when applied to me.
>>
>>>
>>>Some of these people are doing just as well, if not better, than the
>>>guys working away at home and selling their programs. IMO there is not a
>>>lot of difference between academic commercials and commercial
>>>commercials.
>>
>>
>>for the "difference" just ftp to ftp.cis.uab.edu/pub/hyatt and see what
>>you find there.  "the light will come on" and illuminate the difference.
>>Then read rec.games.chess.computer and this message board, and watch at
>>who gives real details and data.  Again the difference will be apparent.
>
>First you say you're not an 'academic commercial', then you say to see
>the difference between 'ac' and 'cc' come and visit your site. Well
>since you're no 'cc'; this presumably places you in the 'ac' camp, no ?
>
>Finally, are you seriously trying to imply that there is no academic
>career benefit from your computer chess work and from articles published
>?
>

none that I have seen, no.  In my academic circles, success is measured
in
research dollars.  Chess has produced *0.00*...

And no, I don't buy the AC label *period*.  Perhaps Berliner with
Federal
money to develop HiTech fits.  But that isn't me.



>I mean professors are supposed to go out and do stuff and create
>opportunities and get stuff published, and have research students, no ?
>Don't the bean counters add up the numbers of words each year, and so on


Nope.  Never had a student working on Crafty nor Cray Blitz.  Never was
the
subject of a grant proposal.  Never was the subject of a course.  Just a
fun
"hobby" of almost 30 years...


>?
>
>Nothing wrong with it. Its a well known process.
>
>Anyway. Why not try and arrange nuclear armaments reduction talks with
>all the other usual culprits; and we could all go back down to K233's ?
>
>Chris



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

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