> They just decided to peg the list to Deep Shredder 12 single core being 2800. This is to make the top engine ratings realistic in human terms
Is there any proof that Shredder 12 x64 1 core should be 2800 Elo (equals to Human Elo) on AMD 4600 2.40 GHz ?
In my estimation,Chess Tiger 2007.1 Elo Performance should be around 2750 Elo Human points (on QX9650 @3.80 GHz)
Probably SCCT Auto232 Elo calculations will be fixed to Chess Tiger 2007.1=2750 ELO:
BTW,here are available some useful Elo data played by Engines and Humans:
Note:There are no many games played,but however i think Rebel's side is a quite good indicator about the estimations between Engines and Human Elo points
And in my opinion,the engine's Elo performance should be based on Humans Elo points and on hardware processor speeds
As we know,any engine Elo rating depends on hardware speed, opening book, ponder off/on, time control...
One thing more,in reality (on latest decent processors) Chess Tiger 2007.1 is not 2550 Elo points
I expect Chess Tiger 2007.1's real Elo rating should be at least 2700 Elo (in human terms)
Note:i mean those calculations (where ChessTiger 2007.1 is rated at 2550 Elo) are wrong, which are based on Shredder 12 1c's 2800 Elo
But however, CEGT, SWCR, Clemens (which are based on Shredder 12 1c's 2800 Elo) are doing great job -BIG thanks for their works,efforts... !
-All engine ratings should be concentrated (based) on Human Elo points
Otherwise there will be misunderstandings....
I think its meaningless,if the engines which are on the same level as GM of 2700-2800 Elo points to be published approx.150-200 Elo points less (weaker) than the reality
Btw,another useful comparison table- Man vs Machine:
For more details:
Even the current Playchess Engine Elo calculation program is too low,as we see many Top engines (Houdini,Rybka...) are rated around 2400-2800 Elo
But in reality (i mean for Playchess engine ratings) the top engines real Elo performance should be rated at least 400-600 Elo higher
That's why i use more accurate starting elo calculation:3200
For example,the bellow engine ratings are more equal to Human Elo points:
Rank Name Elo + – games score oppo. draws
315 Amd64bit, Shredder 7 2662 48 54 640 4% 3230 3%
316 David-Steiert, Junior 9 2632 75 86 232 3% 3215 2%
I mean the current Top MP engines (e.g Rybka,Houdini,Critter...on latest decent i7 6 core machines) play approx. 500-600 Elo stronger than the Top GM of 2700-2800 Elo Human points
Btw,it seems Playchess Elo calculation program is too old-dated (needs update)
Probably Playchess Elo calculation program is based on 2000 years
Normally in those days was ok with the calculations..., as we remember the top chess programs were on level around 2600 Elo
But nowadays,the speed of the processors are changed,the strength of the engines are changed too
And i hope/think the Playchess Engine Elo calculation method should be changed (updated) too
>But what is your evidence that any engine would rate 3350 -3400 Elo against humans?
Sure i mean not for every engine,but i expect (especially at blitz) some of the Top Mp Engines (e.g Houdini,Rybka,Critter) would rate 3350 -3400 against Humans
Just i'd like to mention again that there are 2 (two) main important factors,which is required for high elo engine performance:
-very strong opening book and very fast hardware
>Carlsen, Aronian, Kramnik, and Anand would score about 4% against them at 40/2 hours
We know that Kramnik already lost against Deep Fritz in 2006 and the result was:4 draws, 2 wins (in favor for Fritz)
Really i wonder a lot about what will be the Elo performance of Houdini,Rybka,Critter against Carlsen ?!
Personally i expect to see at least 500 Elo difference (in favor for machines)
I wonder also,why in the latest 5-6 years,there is no any serious match - Man vs Machine ?
Is that can be the reason that there will be a lot of differences in points-between human vs computer ?
>Joel Benjamin did it 25% of the time against Rybka
Yes...Joel Benjamin is a strong anti-program master
And as far as i remember,Joel played against older Rybka version (i think the version was Rybka 2.3.2)
Note also that the current Rybka 4.1 mp is approx.150-200 Elo points stronger than Rybka 2.3.2 mp
Probably Rybka Cluster + superior opening book will be performed at least 800-1000 Elo better
Btw,i know very well too that for more accurate rating and better conclusion is needed many games,
but anyway if you check more carefully the bellow crosstable and my notes (based on 36 games) you will notice what i mean exactly
At longer time controls the iPhone/PPC version has gained 2 GM norms in Human tournaments and a 2931 rating, 2 years ago. The iPhone version of Hiarcs is still the best on this device.
I really doubt we will ever get a reliable rating of the top engines v Humans. The top engines will probably all rate the same or the opening book will decide the result.
The most interesting test would be a top GM using Computer help v a Cluster engine.
> The most interesting test would be a top GM using Computer help v a Cluster engine.
I disagree, GMs are worse at using computer assistance than other expert Freestylers that aren't GMs. As the computer strength increases, the importance of your OTB strength diminishes and other elements like understanding what positions some engines excel at or are weak in (to avoid them) increases.
Have you seen the Freestyle tournament being run at Infinity Chess?
Any top GM that you can name that is doing well at it?
To get a GM of that level to get out of bed will need several 1000$ To play a match will probaly mean 10's of 1000's. Find a sponsor I will get you a GM. To get a 2800 GM you are talking 100's of 1000's.
My idea was 2700+ GM + Top commercially available hardware v a Cluster.
Later we had quotes from Kasparov indicating that chess skills might not be of paramount importance in Advanced Chess. We also had a few top GMs at Freestyle events. Nakamura was pretty terrible as I recall, and he's known for his computer savviness.
As far as money is concerned, of course you wouldn't expect top GMs to compete for a few thousand dollars, but this doesn't mean that top GMs would dominate if the prize funds were a few orders of magnitude higher.
Two obvious reasons why GM's didn't do very well in past freestyle tournaments: 1) bad hardware; 2) bad opening book
If hardware and opening book are equivalent, I just don't believe that anyone here on the forum will be better than a good GM in freestyle chess.
> Which top GM's played in these tournaments?
If my memory serves me, Jobava was the highest ranked player to participate in Freestyle tournaments.
> I said a top GM 2700+ they are now using computers in ways you could not imagine.
Really? Example please.
have you noticed you make a lot of claims w/o backing them up with one shred of evidence or info? You just got through insulting Uly saying how little he knows and how much you know. When we ask you about your knowledge though, you can never discuss it because it's always top secret.
My mistake for asking you a question again. Won't happen again I hope.
> I said a top GM 2700+ they are now using computers in ways you could not imagine.
If I may ask are you taking ways they can travel with there analysis? (which I have heard a few interesting ideas in this aspect) Or just new different ways to analyze there chess games. And if its the latter is there any benefit to the common corr chess player?
As to how a 2700+ using strong h/w would fare against me using equal h/w: there's no info to have a factual discussion. Only assumptions.
These guys are super competitive, super smart. If the purse was enough to get such a GM playing (and putting his ego on the line) - they'd probably seriously prepare.
I'd wager a 10 game match (me + strong h/w vs. 2700+ using same h/w) would end in all draws or at most +1, can't say to which side...
Of course, I can't see, even with $20K purse, an 2700+ GM having any interest in crossing swords with an unrated patzer... they have nothing to gain.
>Thanks for the compliment Uly.
Might as well have been complimented by Beavis.
>I'd wager a 10 game match (me + strong h/w vs. 2700+ using same h/w) would end in all draws or at most +1, can't say to which side...
Of all the stupid and equally arrogant statements I have ever heard... this has to rank right up there in the top 10.
Harvey why you trying to reason with tweedledee and tweedledum? Just to get a good laugh?
> Of all the stupid and equally arrogant statements I have ever heard... this has to rank right up there in the top 10.
This statement doesn't bode well for an analysis of your intelligence.
As far as HEM is concerned, it is clear that he is a master of the advanced chess domain. So in a no holds barred match between advanced chess teams, I think you'd be perfectly justified in betting on him, remembering that OTB opinion is only one ingredient in the mix, and not one that can't be purchased...
"some groups showed a very high level, arguably, the highest ever seen in any chess tournament. Which prompted the early elimination of known players such as: Mark Eldridge, Ralf Greweling, Alvin Alcalá (current FICGS freestyle champion), Erdogan Günes (long time computer chess expert and author of the Rondo book for the 2010 WCCC in Kanazawa), Clay Hofmeister (winner of the the Mundial Chess tournament in 2010), Patrik Schoupal (EtaoinShrdlu), Mark Noble (New Zealand correspondence chess champion in several occasions), Roger Zibell (Houdini book author) and Herbert Kruse (Kreuzfahrtshiff). And this isn't the only proof, the fact that the first and second placed players in the last Freestyle have, so far, drawn all of their games against centaurs, is something to note."
Personally, I think the GMs would want to avoid such tournaments for fear of being humiliated.
If this were the case, one would expect an unassisted engine to perform at least as well as a centaur, unless you are arguing that people do a better job at managing the clock. Putting that aside though, there are always going to be positions where engines rate two or more moves that result in different games equally. I still believe that in these not infrequent occurrences, the engine method of resolution (random selection) will be inferior to the preference of a strong chess player.
> If this were the case, one would expect an unassisted engine to perform at least as well as a centaur
Not necessarily; I wouldn't take it that far. But I would expect that centaurs are not significantly stronger than unassisted engines.
> Putting that aside though, there are always going to be positions where engines rate two or more moves that result in different games equally. I still believe that in these not > infrequent occurrences, the engine method of resolution (random selection) will be inferior to the preference of a strong chess player.
Generally yes, although I am not sure how significant the effects of this superiority really are. In freestyle chess, the problem is that, unless you have two equally good computers, it will take time to even determine that the engine evaluates two moves (close to) equally. The centaur thus risks ending up playing worse than the unassisted engine which simply has more time to calculate.
The consensus a few years ago was 150 Elo. This was expected to decline over time. I suspect that its still 100 Elo or so, which would qualify as significant in my book!
In freestyle chess, the problem is that, unless you have two equally good computers, it will take time to even determine that the engine evaluates two moves (close to) equally. The centaur thus risks ending up playing worse than the unassisted engine which simply has more time to calculate.
Many of the top teams have pretty impressive hardware resources, so concluding that several moves are seen as equal (or roughly equal) by the engines is certainly an already familiar situation. Each team will have a method of dealing with this. I would expect that if the team had a high quality GM team member, he/she would make the call.
Please note that in most cases (at least non-cluster cases), the centaur teams will be using a lot more hardware than the pure engine. For this reason, I don't think your last comment is generally valid.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill