diep wrote: Nick C wrote:
He quoted it as the reason here:
http://www.hiarcs.net/forums/viewtopic. ... c&start=15
“The pro is that more programmers might be attracted. Paderborn was cancelled
this year. One of the reasons was that there were only 5 or 6 programs that
wanted to play. The organizers asked for reasons and many told that there is
no point getting killed by Rybka running on 40 cores.
“The disadvantage of this is that it will hinder innovation on multi core
machines and on clusters.”
Yes he was quoting someone else, but without this quote, he has no reason to even think about limiting hardware. Interesting that none of these programmers have come forwards.
This is just total untrue. The reason why it there were so little particiants was:
a) they keep it at a date where more than a year ago already ex-participants said from: "we prefer to be with our family instead of play at that date between Christmas and the new year". Some still showed up in 2007, but said: "in 2008 i will not show up if it gets organized at the same date".
b) None of the participants said something about hardware. Note that rybka never has had the fastest hardware in paderborn.
c) most computerchess organizers announce events WAY WAY too late. For a blitz event (international hypercube blitz tournament - 150-200 participants) i'm busy with now i'm already organizing way into 2010 now.
You should drop the word 'rybka'. This is not about rybka.
And forget 40 cores, it's 64 cores the machines that intel soon releases, and i bet that the Hiarcs team soon has a box like that also to connect to.
The processor, still under NDA, most likely is the beckton. It's 8 cores that split into 16 logical cores and just works MP, which means it only works in 4 socket mainboards. I'm not sure what clock it runs at, that's still under NDA.
Intel will demonstrate it all at an upcoming event together with other hardware.
The question i like to ask to everyone here, do you want the chessprogrammers to pay more money, because of some decision taken by Levy alone?
to Mike: there is way more factors you forget to consider. If you limit the number of cores, so if you do not standardize the hardware, then that just costs MORE money, because high clocked cpu's are the MOST EXPENSIVE cpu's.
Highest clocked production cpu is power6 at 4.9Ghz.
It eats about 6000 watt for 1 big node and for a $100k or so you can get a box with a few cores of it.
Note that hardware is not most important thing in computerchess nowadays. The impact of a good book is for example way more than whether you run on 8 cores or 800 cores. Hardware matters only when the differences between 2 programs are tiny. Especially it matters a lot when you let an engine play against itself (the so called "incest test" to quote Johan de Koning - not my words).
Yet of course everyone tries to do his best in every aspect of the game that influences the game. So you try to show up with good hardware.
For the nerds among you: reason why high clocked stuff is so expensive is because there is a relation O ( m ^ 3 ). So to get 1 Ghz higher is really very expensive as you have to put it to the power 3 with respect to power through the cpu and all kind of problems that start to exist (leakage etc). I'm not an expert enough there to explain that to you in a clear manner.
This is why we do not have 10Ghz production cpu's.
The thing that is very bad in limiting the total number of cores is
a) the timing, a few months before the world champs we suddenly get the email that the board being Levy, decided to limit things to 8 cores. Just when people have put MONEY and EFFORT in parallellizing their software for machines with more cores and/or other type of hardware.
but most important is
b) marketing. we get MORE attention as a worldchampionship when there is big hardware. TV, radio is suddenly interested if it is open hardware and shows up. In 2000 in London, the last "single processor microworldchampionship", there was NO ONE from the press for the computerchess.
Who is gonna look for a few nerds who play at retarded overclocked machines?
c) the equipment used to overclock is something you do NOT have at home. So believing that the hardware that actually shows up there is something you can buy in a shop is nonsense. Only some very big fanatics, who work for the government, they manage to achieve what is gonna show up there. So a limitation just means that the amateur engines like Sjeng and Jonny get hurt, it doesn't limit Hiarcs/Rybka.
d) the icga board is technoidiots. they never managed to correctly check hardware. I remember 2000 all too well. Back then it was still relative easy for a person to check. Now it is 2009, it is not so easy now. I would be very surprised if the board suddenly could.
e) As of today we do not even have a definition of what a 'core' is. And whether the box must be inside the tournament hall or is allowed to be remote. Is a 4 socket beckton machine allowed if you say you just use 8 cores out of it?
Is 8 fpga's allowed?
The tournament already happens in a few months time.
You guys realize that only teams with big sponsors can arrange a machine according to specs at last moment?
f) The ones who get impacted most by the new rules are the amateurs.
Now they must buy a 8 core box, next year a 16 core box?
Comeon, only the teams with sponsors can switch so easily.
What amateurs dislike most from events is when just operators show up, as operators ONLY show up if they can kick butt with an engine. There is very FEW operators who show EVER up when they cannot win. Now with Diep of course i manage to kick operators. but most amateurs cannot. What they are sick of, that is operators who get for free $1500 from icga for showing up there.
In 2008 world champs i understood there was at least 3 operators not from China to operate strong engines (Rybka, Hiarcs, Toga). I'm very sure without that $1500 for free odds would've been big that at least 2 out of 3 would not have shown up.