Chess computers amp nodes by seconds

  

Back to forum


Thibault de Vassal    (2009-01-24)
Chess computers & nodes by seconds

What about a thread on chess dedicated computers... dual-core, quad-core or even 10 years old tired processor ?

It could be interesting to gather some informations here on hardware evolution by comparing the speed of chess engines [nodes by second], specifying the operating system & the program that runs the engines (Arena, Chessbase, Fritz 11 interface & so on...)


Normajean Yates    (2009-01-24 16:53:42)
I would be most intesrested in reading..

.. posts in this thread.

though I am afraid i have nothing to contribute - didn't keep statistical track: until 2006, I used engines for analysis of critical positions of my favourite historical games only.. at up to 5 days / move! [but I didnt keep hardware/NPS records - all I can say is I mostly used whatever was the latest version of rebel - that was until 2006. Until 2002 I ran long analyses with crafty also...]

Also, in my experience the hash size does change the PV, not only the time taken to reach a certain ply [ because if a line/subtree is in hash it gets appended - with eval - so that say at 18-ply you are seeing *some* lines which are actually 28-ply or more - so it affects what gets pruned by alphabeta ... and if the engine has forward pruning it affects that much more..]

But this being a centeaur chess site (as far as chess is concerned) I suppose every chess player here knows this... :-/


Don Groves    (2009-01-24 22:57:28)
Chess computers

I am interested in such a thread.

For the numbers to be useful we need to establish a benchmark for comparison, for example, a single core processor at 1.0 GHz. Has this already been done in the computer chess community (of which I am only a very recent member and know very little)?


Wayne Lowrance    (2009-01-25 07:35:22)
chess computers

take a look here, maybe this is interesting.http://www.jens.tauchclub-krems.at/diverses/Schach/fritz9_benchmarks.html


Thibault de Vassal    (2009-01-25 19:10:55)
20716 Kn/s

Interesting, Wayne ! 20716 Kn/s .. just dreaming :)

The names of the winners (computers) :

Intel QX9775 Yorkfield [8 x 4730], Intel E5450 Harpertown, Intel Dual XEON X5460, Intel Qx9650 OC, Intel Dual XEON X5365, Intel Xeon 5462, Apple Mac Pro, 2 x Quad Xeon x5365 [8 x 3000, 12810 Kn/s]

wow...


Don Groves    (2009-01-25 23:56:13)
Wow, indeed!

Those numbers are 100 to 200 times greater than mine!


Thibault de Vassal    (2009-01-26 00:49:52)
Rybka vs. Fritz 9

I suppose this is normal, as you probably use Rybka 3 (looks like its number of nodes per second doesn't mean anything btw). These tests have been made with Fritz 9, right ?


Don Groves    (2009-01-26 01:52:41)
Not Fritz...

No, my tests are with Hiarcs.


Thibault de Vassal    (2009-01-26 20:07:55)
Hiarcs vs. Rybka

Yeah, the same... the number of nodes is really low with Hiarcs. Anyway, the engine must be the same one than in the tests (so Fritz 9) to be significant.


Normajean Yates    (2009-01-28 09:46:51)
NPS comparison is meaningless..

between different engines. [ of course if you run the same engine on a faster computer you get better NPS and more powerful hw+sw combination, but that is trivial.. :)]

You will find ample discussion on this in the rec.games.chess archives 1994-2002 period.

I thought everyone knew this - I thought at least Thibault knew this! ;) [I still think he does, he is just being modest ;)] - that's why I didnt post this before.... :)


Thibault de Vassal    (2009-01-28 10:46:32)
NPS comparison

The only fact is you know more about all this than me :) Great that you find time to enlight us with these details on chess engines.


Tano-Urayoan Russi Roman    (2009-01-28 18:53:31)
In a someway related topic

In talkchess.com there has been a debate over the advances in the past 10 years, which part had causes greater improvement in computer chess, software or hardware interesting reading altough a bit long. http://www.talkchess.com/forum/viewtopic.php?t=25215


Normajean Yates    (2009-01-28 21:13:52)
re - NPS

Well in the mid-90s some of the fairly good programs were open-source : crafty one of the the strongest open-source programs, and its author Bob Hyatt used to dicuss chess-programming related issues...

Even with closed-sourse proprietory programs, however, the *empirical* fact of NPS being not so relevant can be established statistically by just seeing the results of inter-engine tournaments..

[of couse NPS is not totally relevant! a 1 NPS engine is likely to be quite bad in comparison with a 10000 NPS engine on the same hardware! :) And a 0-NPS engine can only generate moves by using a random number generator ;) [not-necessarily uniformly distributed] - or by tablebase lookup ... but is the present position counted as a node? If yes, then 0 NPS can do nothing at all, not even check legality of move! :)


Thibault de Vassal    (2009-01-28 21:32:48)
NPS, Crafty

True, true... and the discussion at Talkchess is very interesting too. BTW I did not check the code & results of Crafty for a while but I'm curious to know if there has been significant improvements since the start of the Rybka era.


Normajean Yates    (2009-01-29 14:59:36)
re: rybka era v open-source

I've not been keeping in touch with post-2002 developments in chess programming [that is, the literature - I have no energy to try to reverse-engineer closed code, and my skills in that are 15-years out-of-date and out of touch :(]

- Also, I think there are fewer and fewer of comparatively strong enough open-source engines now :(

Which is the strongest *open-souce* engine now, and how does it compare to even rybkas of the 3.1.x generation? I do not know [frankly, since 2002 I am too busy with literature and political activity; even though it looks like I am playing chess all the time ;)]

But I'd be interested in knowing and grateful to everyone who posts info this and related questions...

The questions are of the form of comparative strengths of stongest post-2005 open-source engines versus rybka; and *published* new ideas in chess programming which have been implemented and have been shown to improve engine strength.


Normajean Yates    (2009-01-29 19:58:08)
about crafty development..

it stopped at 20.16 or something I think... [a few years ago] - havent checked - and have a bad backache so dont feel like checking :/


Tano-Urayoan Russi Roman    (2009-01-31 09:19:05)
Mr Yates

Crafty current version is 22.10 and is still developing. Right now Dr. Hyatt is working in improving the paralellization. More info of the engine is found at computer chess forum as the author is a regular poster there.
Some people say that Rybka is a product of open source of Fruit, but that is another debate. Right now there are good open sources engines mainly glaurung, stockfish (glaurung derivative) there are also interesting projects with toga derivatives, cyclone, grapefruit, mad prune, etc.
Again I refer to an interesting discussion in talkchess about the advances in the past 10 years, software or hardware, Dr Hyatt explains that hardware development has been more important than new methods in search or evaluations of chess engines.


Thibault de Vassal    (2009-01-31 15:38:45)
Dr. Hyatt

"Dr Hyatt explains that hardware development has been more important than new methods in search or evaluations of chess engines."

Surprising... maybe in a certain measure.


Normajean Yates    (2009-02-01 10:15:31)
thanks Mr R.-Román, and I am *Ms* Yates.

Mr Russi-Román, I am Ms Yates, not Mr Yates.

I thank Mr Russi-Román for the valuable info and link: (and others too) : I'll update myself whenever I have free time [not only from chess play, but from higher-priority things...]

So, what is the state of the art in noticing drawish nodes (blockade, repitition etc) during search? [remember: 'false positives' are as bad as 'false negatives'] Offhand here is one area where significantly good new heuristics would be much more important than hardware..

Another example: at present, has any program - however highly parallel - delivered mate from a suite of mate-in-60+ troitski positions (NN v P) *without* using tablebases? (ignore the 50-move rule for this question. Anyway this site doesnt have the 50-move rule...) I dont know the answer...


Don Groves    (2009-02-02 00:11:53)
Hardware vs. software in chess

It seems to me that what hardware advances do is allow software algorithms to run more efficiently, which then allows deeper analysis in the same amount of time compared to older hardware.

If we want to know the best chess software, we must play many engine vs. engine games on identical hardware. Presumably then, the winner will also be better on more advanced hardware.

Another thing to consider is that the hardware need not be a general purpose computer, but may be specifically designed to run a certain algorithm. In this case, the engine using that algorithm would have a large advantage in an engine vs. engine contest on that hardware.


Wayne Lowrance    (2009-02-03 18:59:13)
octal vs single core in CC

I am using a Octal computer now for perhaps 6 months, before that a very old single core slow comp. came across a inheritance so I splurged. Prior to purchasing for this octal I had many talks with the experienced folks on the "Rybka forum". I inquired would an octal computer provide better play in CC. The typical response was the main advantage is the octal is faster, but my older slower comp given little more time would play close to identical, (but not 100 % identical). I have found this to be the case. I have compared single core with octal core using my octal computer and find the single core lagged the octal on average only one ply, but got the same answers on the whole. I like my MacPro (2.8 gihz) very much don't get me wrong but the most important chess playing feature is the program. I thought some of you would be interested in my little tidbit....... Wayne


Normajean Yates    (2009-02-04 01:45:47)
re Wayne's point:

Well if the program is not written to use parallelism [or written badly so that it spends too much time synchronising outputs of various cpus] then it wouldnt do very well on octal compared to single-core and might do worse :)

Chess programs unless really horribly written should not have this problem: typically different cpus will analyse different branched, and hash is sliced but from time to time 'pooled' - ie the hash slices are updated from other hash slices...

Some very old RAMs would have problems with multiple cpus trying to access them: but that would be early 1980s at most: just before the invention of the personal computer :)


Don Groves    (2009-02-04 02:10:20)
Normajean's point

This is sorta what I was getting at above -- that hardware advances (other than faster clocking) only help if the chess software is written to make good use of them. Otherwise, as Normajean points out, the software may actually perform worse.


Normajean Yates    (2009-02-05 04:19:20)
Don & I - same pt fr diff directions:)

Don - no use if hardware doesnt 'fit' software [engineer's way of putting it ;)]

me - no use if software doesnt 'fit' hardware [programmer's way of putting it ;)]

(prefix 'retired' to 'engineeer' & 'programmer' )

We are saying the same thing!

Another point: debugging parallel programs is much more tricky than debugging - uh - non-'parallel' programs

I have no engineering experience [1] but I'd venture to guess that a 'parallel' electronic machine of any kind [not necessarily a computer] which has to synchronise signals from various sources, and do it fast, would be much more of a design challenge than a non-'parallel' one...

[1] (well I did assemble my first personal-computer - a BBC[2] acorn - by cannibalising parts from discarded damaged BBC acorns and some borrowed parts - with a scope and a soldering iron - does that count? But that was back in 1983-84... and I had help....)

[2] yes, BBC = beeb (British Broadcasting Corporation) - they *used* to do meaningful things - but that was long ago....


Don Groves    (2009-02-06 04:34:00)
Normajean's point revisited

Yes, of course that counts ;-)

I should change my bio since I think "programmer" is more accurate than "software engineer." Although I did build my software with the same care and planning as an engineer building a bridge, so maybe... Who knows ;-)


Normajean Yates    (2009-02-06 18:32:02)
I should change my profile to...

'a little bit of this, a little bit of that' :)

Or using the idiom of days long before I was born: 'I explored various activities and found my limitations in all of them rather fast' :)


Nadia Kaif    (2009-03-20 05:11:08)
Nice idea

Yes a technological inside will be interesting