WCCC 2009 - Limit of 8 cores

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Spock

Re: WCCC 2009 - Limit of 8 cores

Post by Spock » Mon Dec 22, 2008 7:14 pm

playjunior wrote:I wonder who would provide all the engine authors who participate a 40-core machine?

This is not about removing incentives for implementing good parallel search, it just gives a chance to amateurs who don't have 40-core clusters or University chair and funding with 500 core machines to compete. There are 1 billion such regulations in sports like Formula1, so that the rich teams cannot just outspend the poorer ones in a way that the drivers won't matter at all. Here, cars translate to hardware and pilots are the software. Well almost.


And, I think that if you have a really good parallel search compared to competitors, 8 cores would give quite some advantage. Look at scaling difference of Zappa vs engines of the same time.

Analogies are rarely useful in debates like this. Each side chooses an analogy to try and demonstrate that their argument is the right one, and we end up no further ahead. Either there are analogies to fit both sides of the argument, or they actually aren't a good match.

My only reservation is that it can turn out into a "haves" and "have-nots" competition. Not everyone either has the money for the really big hardware or the contacts to obtain the use of it at perhaps no cost.

But the assumption is that everyone can make equally good use of something like a 40-core cluster, which is far from the truth. No, this whole thing is instead all about penalising Rybka. If it was a 2500 rated engine that was using 40 cores, no-one would be in the slightest bit interested. Indeed no-one got terribly excited about the Toga cluster before. It is the fact that the strongest engine in the world happens to have access to it, that is what started this whole thing. It is anti-Rybka.

User avatar
George Tsavdaris
Posts: 1625
Joined: Thu Mar 09, 2006 11:35 am

Re: WCCC 2009 - Limit of 8 cores

Post by George Tsavdaris » Mon Dec 22, 2008 9:41 pm

playjunior wrote:I wonder who would provide all the engine authors who participate a 40-core machine?
It doesn't matter.
WCCC(World Computer Chess Championship) is about bringing your absolutely best. Your absolutely superior configuration. Of engine, book, and hardware and whatever else.

If you don't have enough money to have a 40 or 80 or 500 core configuration, then...., well that's life and we know it. But that shouldn't restrict others and not allow them to have this and that.

For example R.Hyatt(Crafty) has access to a huge 500+ core machine. And can play thousand of games each day to improve the engine itself as also the opening book if he wants it by finding good lines or rejecting bad lines, MUCH faster than others who don't have such superior hardware.

Does this mean that if he wanted to enter the WCCC, he is not allowed to do, because he created the engine AND the book with astronomical(a word ICGA used) hardware that others don't have? This is more than obviously ridiculous.

And don't tell me it's not the same and hardware is much more important that the engine and the opening book, because that would be ridiculous again.
For example a good opening book is worth i believe much more than going from 8 to 16 cores.
To go from 8 to 40 cores i guess hardware is more important than even a good opening book, but effective use of 40 cores is difficult and programmers has to spend a lot of time for having a good effective implementation to take advantage of the 40 cores.
And because they will spend all this time for doing that, to create a program that will run on many cores to play Chess in as higher quality they can do, i think they should get some reward, right? :-D

Instead ICGA with what it decided, lowers the quality of Chess programs can play. :-(
After his son's birth they've asked him:
"Is it a boy or girl?"
YES! He replied.....

User avatar
Harvey Williamson
Posts: 1845
Joined: Sun May 25, 2008 9:12 pm
Location: Media City, UK
Contact:

Re: WCCC 2009 - Limit of 8 cores

Post by Harvey Williamson » Mon Dec 22, 2008 9:46 pm

George Tsavdaris wrote:[

Instead ICGA with what it decided, lowers the quality of Chess programs can play. :-(
But in having to optimise for the 8 core limit maybe they will improve the software that you and I can run at home on our dual quad or 8 core.

User avatar
George Tsavdaris
Posts: 1625
Joined: Thu Mar 09, 2006 11:35 am

Re: WCCC 2009 - Limit of 8 cores

Post by George Tsavdaris » Mon Dec 22, 2008 10:01 pm

Harvey Williamson wrote:
George Tsavdaris wrote:[

Instead ICGA with what it decided, lowers the quality of Chess programs can play. :-(
But in having to optimise for the 8 core limit maybe they will improve the software that you and I can run at home on our dual quad or 8 core.
And why ICGA should care about that?
After his son's birth they've asked him:
"Is it a boy or girl?"
YES! He replied.....

bob
Posts: 20912
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: WCCC 2009 - Limit of 8 cores

Post by bob » Mon Dec 22, 2008 10:15 pm

George Tsavdaris wrote:
Harvey Williamson wrote:
George Tsavdaris wrote:[

Instead ICGA with what it decided, lowers the quality of Chess programs can play. :-(
But in having to optimise for the 8 core limit maybe they will improve the software that you and I can run at home on our dual quad or 8 core.
And why ICGA should care about that?
And I don't buy the argument anyway, because there is SMP search, and distributed search, and possibly (not yet done) SMP + distributed search (I would discount Rybka's cluster mode here as not being very effective at all). Anything done to improve a program's SMP mode will apply whether it is for 2 processors or 200. Indeed, most don't realize how bad their parallel search is, until they try to run on 8, 16 or 32 cores. Then when they start to address the poor performance on larger number of procsesors, the program also benefits on 2 or 4 or 8 processors...

Gian-Carlo Pascutto
Posts: 1196
Joined: Sat Dec 13, 2008 6:00 pm
Contact:

Re: WCCC 2009 - Limit of 8 cores

Post by Gian-Carlo Pascutto » Tue Dec 23, 2008 9:24 am

playjunior wrote:I wonder who would provide all the engine authors who participate a 40-core machine?

This is not about removing incentives for implementing good parallel search, it just gives a chance to amateurs who don't have 40-core clusters or University chair and funding with 500 core machines to compete. There are 1 billion such regulations in sports like Formula1, so that the rich teams cannot just outspend the poorer ones in a way that the drivers won't matter at all. Here, cars translate to hardware and pilots are the software. Well almost.


And, I think that if you have a really good parallel search compared to competitors, 8 cores would give quite some advantage. Look at scaling difference of Zappa vs engines of the same time.
This is provided everybody can afford a fast 8 core machine. This isn't the case, so your post makes no sense at all entirely. It's also (much) more difficult to make good use of non-shared memory system than to make use of an SMP machine.

playjunior
Posts: 338
Joined: Thu Jun 21, 2007 10:53 pm

Re: WCCC 2009 - Limit of 8 cores

Post by playjunior » Tue Dec 23, 2008 10:29 am

And I don't buy the argument anyway, because there is SMP search, and distributed search, and possibly (not yet done) SMP + distributed search (I would discount Rybka's cluster mode here as not being very effective at all). Anything done to improve a program's SMP mode will apply whether it is for 2 processors or 200. Indeed, most don't realize how bad their parallel search is, until they try to run on 8, 16 or 32 cores. Then when they start to address the poor performance on larger number of procsesors, the program also benefits on 2 or 4 or 8 processors...


Right. You buy everyone a 500 core machine-they will come up with quite a good search, for sure.
Or-you optimize Crafty on 500 cores, and come up with a very good parallel search. They have only 2-4 cores at home and a bad search. Let it be so, then Crafty enjoys certain advantage on 8 cores, right?

The enormous scaling difference between Zappa and other (top) engines clearly shows that this holds. You can optimize on your ultra-expensive machine, and enjoy sizable advantage on a down-to-earth machine.

I mean, do you agree that your, Diep's, Cozzie's (and so on) hardware is complete unaffordable for people who do not have University/big corporation/shaikh financing?

Having that hardware gives you big advantage already in design of your software. This is fair I believe. But then, someone is so brilliant that, despite not having your resources, comes up with something good, why do you want to deprive him any chance he has by just playing on overwhelmingly superior hardware?

playjunior
Posts: 338
Joined: Thu Jun 21, 2007 10:53 pm

Re: WCCC 2009 - Limit of 8 cores

Post by playjunior » Tue Dec 23, 2008 10:46 am

Gian-Carlo Pascutto wrote:
playjunior wrote:I wonder who would provide all the engine authors who participate a 40-core machine?

This is not about removing incentives for implementing good parallel search, it just gives a chance to amateurs who don't have 40-core clusters or University chair and funding with 500 core machines to compete. There are 1 billion such regulations in sports like Formula1, so that the rich teams cannot just outspend the poorer ones in a way that the drivers won't matter at all. Here, cars translate to hardware and pilots are the software. Well almost.


And, I think that if you have a really good parallel search compared to competitors, 8 cores would give quite some advantage. Look at scaling difference of Zappa vs engines of the same time.
This is provided everybody can afford a fast 8 core machine. This isn't the case, so your post makes no sense at all entirely. It's also (much) more difficult to make good use of non-shared memory system than to make use of an SMP machine.

Does it make sense if someone cannot afford a fast 8 core machine, but can afford a fast 4-core machine? And has a better program which can actually is able to beat many others playing 8 cores, but would have hard time against 512 cores? I think that makes sense.

playjunior
Posts: 338
Joined: Thu Jun 21, 2007 10:53 pm

Re: WCCC 2009 - Limit of 8 cores

Post by playjunior » Tue Dec 23, 2008 1:25 pm

Btw: when I say "500 cores" its figurative speaking meaning "you have practically unlimited resources". I know that there are architecture differences between "512 core" and "8 core" machines.

bob
Posts: 20912
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: WCCC 2009 - Limit of 8 cores

Post by bob » Tue Dec 23, 2008 2:29 pm

playjunior wrote:
And I don't buy the argument anyway, because there is SMP search, and distributed search, and possibly (not yet done) SMP + distributed search (I would discount Rybka's cluster mode here as not being very effective at all). Anything done to improve a program's SMP mode will apply whether it is for 2 processors or 200. Indeed, most don't realize how bad their parallel search is, until they try to run on 8, 16 or 32 cores. Then when they start to address the poor performance on larger number of procsesors, the program also benefits on 2 or 4 or 8 processors...


Right. You buy everyone a 500 core machine-they will come up with quite a good search, for sure.
Or-you optimize Crafty on 500 cores, and come up with a very good parallel search. They have only 2-4 cores at home and a bad search. Let it be so, then Crafty enjoys certain advantage on 8 cores, right?
Yes..

The enormous scaling difference between Zappa and other (top) engines clearly shows that this holds. You can optimize on your ultra-expensive machine, and enjoy sizable advantage on a down-to-earth machine.
What is this "enormous scaling difference" you are talking about? On 8 cores my speedup averages about 6x. The best you can do is 8x. That is not "enormous" so I am not sure what you are referring to.

I mean, do you agree that your, Diep's, Cozzie's (and so on) hardware is complete unaffordable for people who do not have University/big corporation/shaikh financing?
No. There are ways to access a cluster from directly contacting industry (which I did to become affiliated with Cray in the old Cray Blitz days) or AMD/Intel. Or one can apply for time on public-funded clusters. Etc.. Where there's a will, and some effort, there's a way...

Having that hardware gives you big advantage already in design of your software. This is fair I believe. But then, someone is so brilliant that, despite not having your resources, comes up with something good, why do you want to deprive him any chance he has by just playing on overwhelmingly superior hardware?
Because that is the stated purpose of the WCCC. Again, for the Nth time, you are free to hold any sort of tournament you want. Uniform platform. Uniform operating system. Uniform compiler. Common GUI. Common book. No book. Whatever you want. But the WCCC's sole purpose is to encourage development by having "the best of the best" get together to measure up against each other, no holds barred...

Post Reply