Author: Christophe Theron
Date: 10:38:00 12/14/99
Go up one level in this thread
On December 13, 1999 at 10:00:35, Robert Hyatt wrote:
>On December 13, 1999 at 01:53:02, Christophe Theron wrote:
>
>>On December 12, 1999 at 20:55:15, Robert Hyatt wrote:
>>
>>>On December 12, 1999 at 19:07:26, Christophe Theron wrote:
>>>
>>>>On December 12, 1999 at 10:10:58, Robert Hyatt wrote:
>>>>
>>
>>(snip)
>>
>>>>>gets the move from the server, the time starts. It is now up to the program
>>>>>to read the input, act on it, and produce a move. Remember, "crafty" is 10
>>>>>years 'behind' the commercial programs, so I see no point in telling them how
>>>>>to fix such problems. :)
>>>>>
>>>>>But they _can_ be fixed. Crafty can play a 60 move game in one second if you
>>>>>want to see something _really_ fast.. :)
>>>>>
>>>>>Bob
>>>>
>>>>
>>>>If you want to quote me, Bob, at least do it correctly.
>>>>
>>>>Reminder:
>>>>* You attacked commercial programmers in several ways, you even said they were
>>>>stealing things in Crafty. I think they don't need to.
>>>
>>>First, I _never_ said commercial programmers "steal" things from crafty. How
>>>would you steal something that is given to you? However, I can likely point
>>>out several ideas from Cray Blitz (and now Crafty) that are in _your_ program
>>>as well as others...
>>
>>
>>Go ahead and tell me what ideas from Cray Blitz and Crafty are in my program?
>>
>>Something that you have invented yourself, of course.
>>
>>
>
>I'll name one everyone can confirm. When your score drops significantly, and
>you run out of time in the normal search allocation, do you stop and play the
>move that you know has turned sour, or do you search longer in an effort to
>try to fix the problem? (sometimes called a fail-low time extension). So far
>as I know, "blitz" (prior to cray blitz) was the first program to do this, as
>I watched too many programmers at the 1975 and 1976 ACM events sitting in a
>panic wondering if they would find a better move before time ran out. This was
>written up in an old JICCA, and so far as I know, _everybody_ now does it.
Come on Bob. This is an elementary idea I have discovered myself since years.
Between 1982 and 1994 I had access to almost no information about chess
programming, and I have found myself a lot of things like this.
I would not deny you have been creative here, but I think we have all been on
such things.
> Do
>you take extra time on the first non-book move? I used to and this was also in
>the same JICCA article.
I don't do that. I must admit I have tried after you talked about it here, but I
dropped the idea as it was doing nothing good for me.
I wonder if others have had the same experience with this idea?
> Do you use the "PVS" search? Murray Campbell and I
>talked about this, and implemented it in a backup version of "blitz" in the late
>70's. I then ran this in an ACM tournament game when I had to switch to the
>backup version when the primary machine went down. Several watched the game
>and asked about all the fail highs which were unknown back then. I would hardly
>say it was 100% my idea, as we hatched it up during the usual round-table chess
>discussions. I definitely implemented it first and then explained how it was
>working.
Are you claiming some kind of paternity for PVS/NegaScout?
Here is something I think I'm the only one to do, but I cannot be sure of
course. Maybe you are going to tell me that some guy was doing it in an old ACM
event... I'm using both PVS and alpha-beta in my search. The trick is to detect
where PVS will do better, and where simple alpha-beta is best.
The improvement for this is not huge, but it's definitely better than PVS for
me. It brings probably 50% more of what PVS brought over simple alpha-beta.
>>>>* I said that commercial programmers have been ten times more creative than you
>>>>have been (which is indeed a subjective point of view).
>>>>* I did not said you are 10 years behind. I do not know how many years behind
>>>>you are exactly. But I think you are about 100 elo points behind the best
>>>>commercial programs. That's why you need 3 or 4 times the processing power to
>>>>get close.
>>>
>>>Warped thinking. Lets play a match using a 16 cpu alpha system. Since you
>>>are 'ahead' you ought to crush me, right? I mean, parallel search is not a
>>>creative thing to develop, so you should be able to have one up and going by
>>>tomorrow? I'll contact DEC. :)
>>
>>
>>Creativity is in my opinion also doing new things with common hardware.
>>
>>Why do you need those rare and expensive computers to show your creativity?
>
>
>Your definition of creativity is totally warped. "creativity" means to "do
>something new". Not "to do something new with old hardware".
I have said "Creativity is in my opinion also...". Notice the "also".
> Otherwise we
>would see no creativity in the automobile industry, the electronics industry
>(do you call today's VCR technology "creative"? yes you say? Yet you do know
>that these things first cost way over $1,000 US when they hit the market?)
>
>As for "rare and expensive" do you have any _clue_ how many of these "rare and
>expensive" computers are playing chess on ICC and FICS _right now_? I have
>_two_ students in a class of mine, both with dual-processor "rare and expensive
>machines". You should look around, you might be surprised just how "rare and
>expensive" these things _really_ are...
I'm pretty sure a well designed program running on one processor can beat yours
running on 4 processors.
In this case, "expensive" is the word that comes to mind about your approach.
You may argue that 4 processors boxes are not that expensive (how much do they
cost? 4 times the price of a standard PC?), but you would need 8 processors or
more to get an advantage, and these ARE expensive.
Multiprocessors might be the norm in several years, but I don't think it will be
the norm for the standard PC. You don't need several processors to use a
spreadsheet, an accounting program, or a word processor.
Multiple processors motherboards will always be more expensive than single
processors one. Multiple processors were much more needed several years ago,
when one processor was not enough for many applications.
For example, MPEG decompression cards existed several years ago (for example
Creative MP-400), but now you cannot find them anymore. Some hardware logic in
the graphic cards together with the improved processing power of your main
processor is now enough. There are other examples in sound processing.
Single processors systems will stay the mainstream for long. Internal
parallelization in processors can be simply used to simulate a faster "one head"
processor, and this has begun already (branch prediction, dynamic register
allocation...). Because this is what we need the most. One faster processor is
more useful, is most applications, than 2 average speed processors.
I know that researchs on multiprocessing are useful and will give interesting
results in the future, but my goal is to produce chess programs for the users of
today, using the standard computers of today.
When SMP will be the norm, I'll produce a parallel program. I might also work on
a parallel program to compete on tournaments, but still I think most users are
waiting for improvements on single processor programs.
>>There are plenty of things to do with a single processor.
>
>
>There are plenty of things to do with a dual/quad/etc processor as well...
>single processor machines are the past.
I don't agree at all.
> Dual/quad processors are the
>future. It is likely that within 5 years a microprocessor chip will have
>multiple processors, just like today's micros have multiple pipes.
It's likely that they use parallelization to speed up single thread processing
as well.
>>Reminds me of these children in Africa, India and south America. You would be
>>amazed by their ability to create toys with wood and used cans. And they have a
>>great time playing with them.
>
>And there is nothing wrong with that. But I wouldn't call the folks at Mattel,
>Tonka, Fisher-Price, and the rest "uncreate" just because they are using high-
>tech digital stuff rather than sticks and stones..
Okay, they are. But they are working for the people that can afford to buy these
toys.
>>On the other hands, I see countries where children are unhappy with expensive
>>computer toys, cellular phone toys,... Their biggest pleasure is to destroy
>>them, not to play with them.
>>
>>While you play with your very expensive toys and repeatedly insist that I play
>>with you, let me remind you that almost all computers on this planet are single
>>processor ones.
>>
>
>You had better wake up and look around. "almost all" is far from reality
>now. You only have to check with motherboard manufacturers to discover how
>many dual-processor boards they are shipping. Many are on a 3-6 month back-
>order due to demand. Quad-processor MBs from Intel are _very difficult_ to
>obtain due to backlog. Yes _most_ machines are single-cpu. But a long way from
>_all_ machines...
Reminds me of several things I have heard in the past:
* IA will be used on every computers in 5 years
* Prolog is the language of the future
* ADA is the language of the future
* OS/2 is the OS of the future
I can understand that the demand for SMP computers is growing for specific
domains.
Still, the mainstream computers will still be monoprocessor ones.
>>You give away your program for free, but still show an elitist philosophy about
>>the computer that must be used in order to get acceptable results from it.
>
>Hardly. It runs fine on single-cpu machines. It just happens to run better on
>multiple-cpu machines.
If it was better on single CPU computers, it would be better on multiple CPU
computers...
>>If beating commercial programs by using 8 to 16 times more computing power is
>>what makes you happy, then no doubt you'll stay happy for a while.
>>
>>
>>
>>>Your comment says this: "what I (you) work on is creative. What you (me) works
>>>on is not creative." I say _horsefeathers_. Perhaps after you start to work
>>>on a parallel search, and use all the information I and others have published,
>>>you will be able to 'create' exactly what we have done?
>>
>>
>>We'll see when I work on this problem. So far it's not in my priority list.
>>
>>My goal is to write a good chess program for everybody's computer, not to win by
>>having more computing power. So far single processor is the most common computer
>>platform for my "target".
>>
>>
>>
>> Christophe
>
>
>So far. 10 years ago everyone was saying "32 bit processors are too expensive
>and we simply don't need that much computing power in home computers." Does
>that sound familiar? Does it sound something like what you said above? Check
>back with me in 5-6 years...
I'll check with you.
Christophe
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.