| |
Main
Date: 10 Jan 2009 14:28:25
From: lonewolf
Subject: chess and grid computing
|
I'm wondering if there has been any research or work done on the
problem of chess and grid computing. Specifically, it seems to me
that the "problem" of chess can be decomposed easily enough (by
sending nodes of the search tree to nodes of the network for
evaluation) and theoretically if its then possible to create a chess
playing network that could possibly be quite good (given enough
participants). Imagine a million PC's working together and playing
Hydra? I'd like to see that game.
Thoughts?
|
|
| |
Date: 12 Jan 2009 08:01:02
From: Sanny
Subject: Re: chess and grid computing
|
> > Nvidia claims its graphics card has 200-400 cores and can speedup
> > Applications by a factor of 10-50.
>
> >http://apcmag.com/nvidia_lands_cuda_100x_faster.htm
>
> That'll speed up graphics rendering. =A0It has no effect whatsoever on
> the performance of a chess engine. =A0And I'd be interested to hear how
> going from a single-core to quad-core processor could get you much
> more than a factor of four in performance.
Each processor in a Quad Core process 2 threads at a time So it can
handle 8 threads simulataniously.
Further using Graphics card a few programs can be speed up by a factor
of 10-50.
Bye
Sanny
Play Chess at: http://www.GetClub.com/Chess.html
|
| |
Date: 12 Jan 2009 14:44:19
From: David Richerby
Subject: Re: chess and grid computing
|
lonewolf <[email protected] > wrote:
> I'm wondering if there has been any research or work done on the
> problem of chess and grid computing. Specifically, it seems to me
> that the "problem" of chess can be decomposed easily enough (by
> sending nodes of the search tree to nodes of the network for
> evaluation) and theoretically if its then possible to create a chess
> playing network that could possibly be quite good (given enough
> participants). Imagine a million PC's working together and playing
> Hydra? I'd like to see that game.
Things that distribute or parallelize well are things that can be cut
into small packages. Distribution is great for cracking codes because
you can have each machine try a bunch of different keys. It's great
for finding messages from the aliens because you can give a chunk of
data to each machine and ask if there's a message in there.
Chess doesn't distribute particularly well because of transpositions.
When a single processor machine considers 1.d4 d5 2.c4 it sees the
position it just saw after 1.c4 d5 2.d4 and doesn't need to do any
more work. In a distributed system, you have to do all that work
again because you don't have the database of positions that all the
other nodes have seen.
Even on a multiprocessor machine with shared memory (and, therefore, a
shared database of transpositions), things are difficult because you
have to be very careful about concurrent accesses to the table. This
co-ordination hits the efficiency of the system and it's even worse
in the distributed case. Multiprocessors and distributed systems can
search faster than a single processor but it's nothing like `ten
processors or ten machines is ten times faster than one processor.'
Dave.
--
David Richerby Happy Spoon (TM): it's like a piece
www.chiark.greenend.org.uk/~davidr/ of cutlery that makes your troubles
melt away!
|
| |
Date: 12 Jan 2009 06:41:59
From: Sanny
Subject: Re: chess and grid computing
|
On Jan 12, 1:52=A0pm, Simon Krahnke <[email protected] > wrote:
> * Sanny <[email protected]> (09:14) schrieb:
>
> > With Quad Core processors and Graphics cards you can speed =A0up any
> > application by a factor of 10.
>
> Oh, fantasy land again.
>
> mfg, =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0simon .... l
Nvidia claims its graphics card has 200-400 cores and can speedup
Applications by a factor of 10-50.
http://apcmag.com/nvidia_lands_cuda_100x_faster.htm
World is stranger than fiction.
Bye
Sanny
Play Chess at: http://www.GetClub.com/Chess.html
|
| | |
Date: 12 Jan 2009 15:52:33
From: David Richerby
Subject: Re: chess and grid computing
|
Sanny <[email protected] > wrote:
> Simon Krahnke <[email protected]> wrote:
>> * Sanny <[email protected]> (09:14) schrieb:
>>> With Quad Core processors and Graphics cards you can speed up any
>>> application by a factor of 10.
>>
>> Oh, fantasy land again.
>
> Nvidia claims its graphics card has 200-400 cores and can speedup
> Applications by a factor of 10-50.
>
> http://apcmag.com/nvidia_lands_cuda_100x_faster.htm
That'll speed up graphics rendering. It has no effect whatsoever on
the performance of a chess engine. And I'd be interested to hear how
going from a single-core to quad-core processor could get you much
more than a factor of four in performance.
Dave.
--
David Richerby Psychotic Pants (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ well-tailored pair of trousers but it
wants to kill you!
|
| |
Date: 12 Jan 2009 09:52:27
From: Simon Krahnke
Subject: Re: chess and grid computing
|
* Sanny <[email protected] > (09:14) schrieb:
> With Quad Core processors and Graphics cards you can speed up any
> application by a factor of 10.
Oh, fantasy land again.
mfg, simon .... l
|
| |
Date: 12 Jan 2009 00:14:46
From: Sanny
Subject: Re: chess and grid computing
|
On Jan 11, 3:28=A0am, lonewolf <[email protected] > wrote:
> I'm wondering if there has been any research or work done on the
> problem of chess and grid computing. =A0Specifically, it seems to me
> that the "problem" of chess can be decomposed easily enough (by
> sending nodes of the search tree to nodes of the network for
> evaluation) and theoretically if its then possible to create a chess
> playing network that could possibly be quite good (given enough
> participants). Imagine a million PC's working together and playing
> Hydra? I'd like to see that game.
>
> Thoughts?
I do not understand what do you want to do with Grid Computing?
If you want to play Chess you can play at: http://www.GetClub.com/Chess.htm=
l
Todays computers are so fast that no task takes long time.
With Quad Core processors and Graphics cards you can speed up any
application by a factor of 10.
So speed is not a limit for todays computers.
What do you want to do? Do you have enough budget for Grid Computing?
Bye
Sanny
Play Chess at: http://www.GetClub.com/Chess.html
|
| |
Date: 12 Jan 2009 07:42:34
From: Simon Krahnke
Subject: Re: chess and grid computing
|
* lonewolf <[email protected] > (2009-01-10) schrieb:
> I'm wondering if there has been any research or work done on the
> problem of chess and grid computing. Specifically, it seems to me
> that the "problem" of chess can be decomposed easily enough (by
> sending nodes of the search tree to nodes of the network for
> evaluation)
Evaluating a single position is far too easy to be done to be
distributed. Distribution adds too much overhead.
A grid is like a slowly communicating cluster. SMP systems a least have
shared memory, clusters only have there interconnects and grids got the
Internet.
So you distribution units that are calculated within seconds, which
quite large when a single node in the grid can be a powerful cluster.
> and theoretically if its then possible to create a chess
> playing network that could possibly be quite good (given enough
> participants). Imagine a million PC's working together and playing
> Hydra? I'd like to see that game.
>
> Thoughts?
If it were that easy people wouldn't care to build Hydra.
mfg, simon .... l
|
| |
Date: 11 Jan 2009 21:54:08
From: Tony M
Subject: Re: chess and grid computing
|
On Sat, 10 Jan 2009 14:28:25 -0800 (PST), lonewolf
<[email protected] > wrote:
>I'm wondering if there has been any research or work done on the
>problem of chess and grid computing. Specifically, it seems to me
>that the "problem" of chess can be decomposed easily enough (by
>sending nodes of the search tree to nodes of the network for
>evaluation) and theoretically if its then possible to create a chess
>playing network that could possibly be quite good (given enough
>participants). Imagine a million PC's working together and playing
>Hydra? I'd like to see that game.
>
>Thoughts?
Here's one fairly recent example:
http://www.grappa.univ-lille3.fr/icga/program.php?id=520
|
|
