Regarding ICGA 2009 tournament

[diep]

Hi,

Here is what i had emailed 19th of december to David Levy regarding ICGA new idea of suddenly and unexpected limit to 8 cores.

Good morning,

Some of my considerations already been posted to CCC,
so i would prefer to refer to that.

Obviously i am for open hardware and 1 tournament and 1 world title.
All the other shit we had in the past already and it was very bad,
it only means political games get played not benefitting anyone except the corrupt.

More information:

0) MARKETING. Not even my own chessteam took Diep serious, until it ran at a 1024 processor box
in 2003. Though it was bad prepared that world championship (other than the openingsbook being really
well done by Arturo) as its evaluation had not been improved, and i couldn't test on the box, i felt like
a total beginner there; my teammembers suddenly took diep serious. The years before they never took it
serious. Only Fritz was serious.

Chessplayers understand, just like normal human beings, that if BIG hardware gets used somewhere,
that an event is serious.

This is just so so important and it gets proven over and over again.

It is already difficult to get attention for computerchess now, basically parallel search is a motivation
for most to take the field serious. Making the tournament tinier than it already was, is not a good idea.

The way to get many participants is to be fair towards the participants.

If ICGA demands 40k euro 'organisation costs' then obviously the participants will know this somehow,
as some of them work in financial areas they see a flow of numbers pass their screen and will see anyway
how much or how little ICGA demands.

If there is a clear encouragement like: PROGRAMMERS can get a free ticket and free hotel room.
Not some hotelroom shared by 5 persons or so (like Reykjavik). Note some free ticket announcement
2 weeks in advance but like half a year in advance so that MANY can profit from it and not just the few
who happened to get the email.

DO NOT PAY FOR OPERATORS,, THE BIG MISTAKE OF THE ICGA.

Do not pay for OPERATORS.

PAY OPERATORS NOT A PENNY.

That is the big mistake ICGA makes each time again.

Each time you see a new dude. Who knows nothing, who says nothing, who adds nothing, and who takes
care less programmers (and therefore programs) show up.

Now i can talk on how bad it is to have a 8 core limit 2009 as it would mean i must think in juridical terms
that i invested in buying a network. It would be bad to not allow a $399 videocard with 240 cores to join.
If hydra would want to join, i want to play it. At least invite it please (though we both agree odds are tiny
it will join as it would get butchered).

Whereas this is a lot cheaper than the 8 core skulltrail boxes that N*SA/Google type organisations own,
as it can have 4 videocards (good for number crunching for them, either AMD/ATI or nvidia videocards it can have).
It is impossible to compete with the sysadminst of N*SA with their nitrogen overclocking.
Not only expensive, also really hard to accomplish.

Give the cheapo clusters the opportunity, despite inferior old hardware, to beat expensive machines from N*SA.

What would be interesting is INSPECT the hardware. Even when that is a remote inspection. Just to avoid
NDA'ed machines and all the lies that happen then, which is why we have this 8 core proposal.

The real reason was that last world champs NDA'ed machines join and on the internet we see cover ups now
of it. The search depths shown however cannot be lied about...

And MONITOR the network of the hardware, so that weird connections to the box are not possible.

That would be great improvements.

With the above you've got 30 participants handsdown.

You cannot expect participants to treat an organisation different from how they treat the participants.

If an organisation wants big money and gives back little and what they give back usually to operators,
then do not expect programmers to join. That is reality.

If an operator is really motivated for a specific program, let him pay his own costs. Don't reward operators.

Consider this:
The 8 core limit in 2009 is total silly if a $399 videocard has 240 cores.

The guy who manages to get a chessprogram to work on it real well, let him please PROFIT from that,
as it is so so hard to manage it.

Vincent

[Spock]

Consider this:
The 8 core limit in 2009 is total silly if a $399 videocard has 240 cores.

The guy who manages to get a chessprogram to work on it real well, let him please PROFIT from that,
as it is so so hard to manage it.

Vincent

That is I guess a possible loophole in the 8-core limit. The NVIDIA Tesla computers actually tend to have 8 CPU cores or less - dual Xeons. So you could argue that you are competing on an 8-core machine in total accordance with the rules.

Of course you would most likely be thrown out. The rules say 8 cores, not 8 CPU cores, but then again "cores" are not defined. The videcard architecture is somewhat different. But that is an example of the new rules stifling innovation - no incentive for a chess programmer to get a program working on the nvidia cards.

[Harvey Williamson]

Consider this:
The 8 core limit in 2009 is total silly if a $399 videocard has 240 cores.

The guy who manages to get a chessprogram to work on it real well, let him please PROFIT from that,
as it is so so hard to manage it.

Vincent

That is I guess a possible loophole in the 8-core limit. The NVIDIA Tesla computers actually tend to have 8 CPU cores or less - dual Xeons. So you could argue that you are competing on an 8-core machine in total accordance with the rules.

Of course you would most likely be thrown out. The rules say 8 cores, not 8 CPU cores, but then again "cores" are not defined. The videcard architecture is somewhat different. But that is an example of the new rules stifling innovation - no incentive for a chess programmer to get a program working on the nvidia cards.

There are no loopholes yet as the wording of the 8 core rule is not finalised. Here is an extract from David's 2nd email the full text of which is on the Hiarcs forum http://hiarcs.net/forums/viewtopic.php?t=2008&start=21

Some of those who have analyzed the technical details of the 8-core proposal have pointed out that the original proposal as presented needs a certain amount of clarification, and this will be forthcoming during the coming weeks, after the ICGA takes further recommendations from prospective participants. For example, the rule could be that not more than 8 cores may be used, so if someone has an 8-core machine that hyperthreads to 16 cores it would be necessary for the operator to disable hyperthreading.

Whose opinions should be given the most weight?

We believe that in this debate the views of chess programmers are more important than those of the other groups mentioned above, and in particular those programmers who have participated in the WCCC in recent years.

Taking into account the postings on the Internet and the emails received by the ICGA, we have analyzed the opinions as follows for the authors of those programs that have competed during the past five World Computer Chess Championships. Exactly half are in favour of the idea and half are against. One person in this group would not express a preference. Many of those who have competed during the past five years have not yet provided any comment.

What changes or refinements if any should the ICGA make to the original announcement?

Given the shortage of time before the start of the 2009 event (less than five months) the ICGA will not make any fundamental change to the concept of restricting participants to the use of 8 cores. Some refinements of the rules will be necessary in order to ensure that all participants keep within the spirit of the restriction as well as within the letter.

[Gian-Carlo Pascutto]

Given the shortage of time before the start of the 2009 event (less than five months) the ICGA will not make any fundamental change to the concept of restricting participants to the use of 8 cores. Some refinements of the rules will be necessary in order to ensure that all participants keep within the spirit of the restriction as well as within the letter.

It's really amazing how you can change the rules 5 months before the start of the event, and then declare that protests are useless because it's too short before the event to change the rules. Only Levy can come up with something like that.

Meanwhile, it's still unclear whether the following is allowed:

a) remote hardware (if allowed, there is no way the rules can be enforced)
b) video cards (how to count cores for that?)
c) FPGA

[Spock]

It's really amazing how you can change the rules 5 months before the start of the event, and then declare that protests are useless because it's too short before the event to change the rules.

Agreed.

.

[George Tsavdaris]

It's really amazing how you can change the rules 5 months before the start of the event, and then declare that protests are useless because it's too short before the event to change the rules. Only Levy can come up with something like that.

That was the first thing i noticed also. It's really amazing indeed!!

First they change the rules fundamentally about 5 months before the tournament and without asking any opinion from the programmers, then they ask for programmer's opinions and then they say that whatever opinions programmers will have, it's too late to change the rules back.
Funny stuff.

[chessfurby]

Seconded.

However may i remind you that the only thing the so called "sport officials" (in any sports, be it track and field, swimmin etcetc) are good in is:

-taking and wasting money which is not theirs
-making irrational decisions often not in favour of the athletes/ competitors/ even working against the sports/ sportsmanship

Basically it is not uncommon that they always have an open hand (for taking whatever stuff...) but never an open ear for their athletes. Funny thing is the sole reason for the existence of these sport officials is... to be the voice of sports and to represent thir members/ competitors/ athletes.

You gotta make up your own mind to decide in which category Mister Levy falls...

[diep]

Given the shortage of time before the start of the 2009 event (less than five months) the ICGA will not make any fundamental change to the concept of restricting participants to the use of 8 cores. Some refinements of the rules will be necessary in order to ensure that all participants keep within the spirit of the restriction as well as within the letter.

It's really amazing how you can change the rules 5 months before the start of the event, and then declare that protests are useless because it's too short before the event to change the rules. Only Levy can come up with something like that.

Meanwhile, it's still unclear whether the following is allowed:

a) remote hardware (if allowed, there is no way the rules can be enforced)
b) video cards (how to count cores for that?)
c) FPGA

d) not to mention cheap clusters, a lot cheaper than 8 core machines
overclocked to 8 Ghz are with Nitrogen cooling,
and more useful as you can also test games with it at an efficient manner,
which a single 8 core machine doesn't give.

The proposal as a concept of limitation to 8 cores is just a show of incompetence.

Vincent

[diep]

Given the shortage of time before the start of the 2009 event (less than five months) the ICGA will not make any fundamental change to the concept of restricting participants to the use of 8 cores. Some refinements of the rules will be necessary in order to ensure that all participants keep within the spirit of the restriction as well as within the letter.

It's really amazing how you can change the rules 5 months before the start of the event, and then declare that protests are useless because it's too short before the event to change the rules. Only Levy can come up with something like that.

Meanwhile, it's still unclear whether the following is allowed:

a) remote hardware (if allowed, there is no way the rules can be enforced)
b) video cards (how to count cores for that?)

SNIP

On remote hardware: see one of my previous emails. It's too difficult to check remote hardware to even KNOW how many cores it has, or whether there is some backlink to another machine with more than that number of cores.

On videocards:
Programming these is going to be most interesting as market must take them more and more into account with respect to number crunching.

I would argue a videocard is having as many cores, in fact manycores, as there is 'threadblocks' possible to run on them.

For the non-x2 cards that is in case of Nvidia at the 8800 a threadblock or 8 and i understand that the double GPU versions also have 8 threadblocks.

So i would see that as 8 independant cores. The other cores are in fact manycores which all execute the same instruction. This is similar also to vector processing.

I'm not so sure about the x2 cards how many threadblocks nvidia can start there. In fact x2 cards are not so attractive to chess because of how the memory works.

With respect to AMD/ATI the cards are having more potence, but are also way harder to program because keeping busy in a chessprogram 5 execution units in each of the manycores is going to be really tough.

The whole card is 1 manycore with 64 streamprocessors or 320 execution units (AMD even markets it in latest card as 1600 stream processors, i would call that execution units though, it's in reality 160 stream processors @ 5 execution units and that in a difficult to use x2 version where one half of the RAM is not accessable by the other half, rendering shared hashtables near impossible). So in case of AMD i would see the non-x2 cards for sure as 1 core and probably the x2 cards also as they are put in parallel.

c) FPGA
I would argue 1 fpga card implementing 1 chess search is 1 processor.
Note that FPGA development boards as hydra uses them are really ugly slow compared to todays processors.

It's like a 60Mhz clocked FPGA card that tries to fight a 4+ Ghz core2?

Not to mention the losses in hardware.

So allowing hydra to join at 8 fpga cards/cpu's would be a possibility.

Note the sheikh's been cheap skate so far, paying and doing the minimum to get hydra done; they join at fpga development cards instead of printing a few fpga cpu's, which are much higher clocked.

For the reader: what is faster 64 x 60Mhz development cards or
8 * 600Mz fpga cpu's?

Hah we probably don't want to know, fpga is so so inefficient!

Point is, let hydra join.

d) clusters
very easy: count the total number of cores

note that parallel losses at a cluster are much bigger than at a shared memory machine. This is why shared memory machines are that much more expensive; they are rare and much wanted and very easy to use.

That determines their price.

It is all about the number of individual steerable cores you have in total IMHO.

e) logical cores
A logical core can get independant steered. Whether it gives a speedup or not, same method for them like clusters; if hardware is capable of logical cores, simply count all logical cores it is capable of, not whether you use them or not.

Vincent

p.s. i guess Junior team has quite more important things to do right now (mankind is evil huh?) when they have time, it would be interesting to know their reaction; A few years ago (2001) i remember they were for open hardware and found joining at slower machines big nonsense. Interesting to know their current viewpoint.

[M ANSARI]

All this 8 core stuff came about as a way to try and stop Rybka from winning. The truth of the matter is that Rybka would have won cluster or no cluster ... and if anything, using the cluster was a much riskier venture than using a highly overclocked 8 core, in one case it made a big blunder due to still green implementation. I find it interesting that nobody ever thought of trying to control hardware before when almost every single engine was running single core.

Anyway ... not having big hardware will only diminish interest in the tournament. I know that for me it won't be interesting as I have two 8 cores and I have almost every engine that has a chance of placing in the top and I can make a tourney in my own home. What would interest me, and most other people interested in computer chess, would be a tourney that will knock the socks off anything I have or can get my hands on.