Author: Vincent Diepeveen
Date: 04:38:46 08/04/05
Go up one level in this thread
On August 04, 2005 at 07:00:24, Engin Üstün wrote:
A world champ is a world championship because the BEST play there.
At your own home you can keep 'equal' hardware tournaments.
Please note that Fritz has been written in assembly to run optimal at P4
processors. Though it is faster on AMD, it is optimized for P4.
In the past 'micro' world championships had the major problem that more money
bought bigger machine.
This whereas to get a sponsor for a quad box at a world champs is relative easy
to find.
I have right now 1.8Ghz dual core processors, which will be the SLOWEST core
probably joining the event.
Nevertheless i'm very happy.
If we would organize a single cpu event then
a) i would not get a sponsor nor would anyone else
b) the event would get uninteresting for media to cover
c) the sponsor of the event will put at the event 30 machines of
P4 2.4Ghz and not a 3Ghz watercooled A64 X2.
So i would show up with my dual k7 machine which i own and play there
single cpu and the company with the most
money shows up with some liquid cooled at -100C montecito cpu at 4.2Ghz.
So the reality is that in single cpu events money is MORE important than
in open hardware events.
d) i would perhaps not join, neither would anyone else as no one on the
planet cares about how something performs at outdated P4 2.4Ghz hardware.
People look to formula 1 race cares because they are monsters with
special tuned engines that deliver 900+ horse power and big wheels, yes
a single wheel of such a car is more expensive than my own. The total cost
of a formula 1 car, just the car components added, is between 1 million dollar
and 5 million dollar for the big teams. The total budget for a formula 1 team
is between 30 million dollar for a small team up to 300 million dollar for big
teams a year.
Now THAT is interesting to watch.
I NEVER in my life watched a race with a renault megane 1.6 liters with 110
horse power.
Now we compete in world champs 2005 with machines that have a total price of
just 12000 euro at most. Paid by sponsors. Every strong parallel program can
easily get a sponsor for a dual core dual machine, or a quad dual core machine.
I tried in fact getting a 8 processor dual core, but my sponsor didn't want to
buy that machine. His money, his decision.
It is not my fault that there is so little parallel programs around. I've been
helping many chessprogram authors to setup their parallellism and get it to
work. Really, many authors. Everyone who ships me an email asking for advice
i directly ship back a lot of information.
Next year there is the cell processor which has 8 cores and quad core opterons
are there.
So any single core event is just such an outdated nerd idea, always raised by
people who are single core themselves, that it is pathetic to think about it.
Just parallellize your engine, or face reality that you will have to fight
single cpu against parallel engines.
Personally i prefer fighting against big supercomputer monsters, but i am a
realist. I ran myself at such machines and know that it is unlikely that any
other program other than diep, crafty or hydra will ever run at a supercomputer.
Right now hydra doesn't actually work at a supercomputer nor does crafty,
so odds are real tiny.
Diep does work at them. In fact i can get without problems a cluster for a big
event. But it HAS to generate a lot of publicity. The current format of the
world champs already doesn't deliver enough publicity for me to get big hardware
from some supercomputer manufacturers.
There is a big supercomputer in france. 8000+ processor itanium2. It is by
far the fastest supercomputer on the planet from computerchess viewpoint,
as it has quadrics QM500 cards and 2 of them at each memory node.
At such a supercomputer diep can work up to quite some processors. Note i
would take a node or 32, not more.
This quadrics sponsor i can get for big events, but in a single cpu event
no one is interested. People like to look at dream hardware.
In reality those quad machines are already cheaper than their car. If just 1
other supercomputer joins such event, yes even a pc cluster like hydra would
be enough, then that already generates so so much more publicity, that it
makes it worth joining the event.
What happens now is that even Fritz doesn't join anymore.
Of course they have already their own world title in bilbao, ICGA recognized
(i'm actually pretty pissed about that action), so they won't join the
world champs in 2005 thanks to having their own worldtitle in bilbao already.
Probably the world champs in 2006 in Turino will again generate a lot of
publicity, after which everyone is going to join there.
No one can ignore big publicity events.
But no one cares about a single cpu event.
This where the last few months, all i've been doing is optimize diep for
single cpu. So i have not busy optimizing diep for big hardware.
I just measured whether it ran parallel ok at the quad opteorn dual core,
that's all.
So the work is there to let it play real well at a single cpu, but the reality
is that no one cares for it. They want to see them on dream hardware, just like
they want to see car races with dream cars with immense horsepower.
Vincent
>Leo's pages is very interesting because he made tournaments on same hardware
>with ponder on, on dual core mashine.
>
>fair is to compare the engines on each division on same equal hardware + with
>same hashtables and same time.
>
>In other tournaments, padaborn, WCCC, CCT, this is not to see that the hardware
>was the same, only the time control is the same, that is uninteresting.
>
>interesting is all engines with same maybe 32 core mashine with 1 GB hash and
>same standard time control.
>
>And of course the engines to use all processors (named "Deep" ).
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.