1
2
![[ch]](/mwf/v12/flags/ch.png "Switzerland")
Hi all
I am considering buying a new computer mainly for chess.
I was thinking about buying 2 CPU Xeon x5650 and OC them.
For that I would need an appropriate
motherboard
RAM
CPU cooler (I want to us air cooler)
Power supply.
Do you have recommendation brand/model for each of these ?
More specifically for the motherboard which I don't remember the ones allowing to OC Xeon CPUs.
Thanks for your help
I am considering buying a new computer mainly for chess.
I was thinking about buying 2 CPU Xeon x5650 and OC them.
For that I would need an appropriate
motherboard
RAM
CPU cooler (I want to us air cooler)
Power supply.
Do you have recommendation brand/model for each of these ?
More specifically for the motherboard which I don't remember the ones allowing to OC Xeon CPUs.
Thanks for your help
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
The EVGA Classified SR-2 MoBo seems the most likely candidate and then components to suit. As a guide, see for example ...
http://3xs.scan.co.uk/ConfigureSystem.asp?SystemID=1228
Scan seems to limit O/C to 3.8 GHz
PeterG
http://3xs.scan.co.uk/ConfigureSystem.asp?SystemID=1228
Scan seems to limit O/C to 3.8 GHz
PeterG
That seems to be an interesting supplier.
A 12 cores at 3.8 GHz should be very powerful
A 12 cores at 3.8 GHz should be very powerful
For configuration build guidance they also supply pre built
http://3xs.scan.co.uk/ConfigureSystem.asp?SystemID=1396
I took off the gaming frills from this one, added a hard drive and with 24Gb RAM priced it at £3784 inc VAT that seems a good price.
Based around a workstation there is also
http://3xs.scan.co.uk/ConfigureSystem.asp?SystemID=1250
again, getting rid of the frills plus extra hard drive came out at £3718.75
If you decide to build yourself ( I could not do it cheaper than Scan based on components purchase) be aware the ESR MoBo is back and only limited cases will accommodate it.
PeterG
http://3xs.scan.co.uk/ConfigureSystem.asp?SystemID=1396
I took off the gaming frills from this one, added a hard drive and with 24Gb RAM priced it at £3784 inc VAT that seems a good price.
Based around a workstation there is also
http://3xs.scan.co.uk/ConfigureSystem.asp?SystemID=1250
again, getting rid of the frills plus extra hard drive came out at £3718.75
If you decide to build yourself ( I could not do it cheaper than Scan based on components purchase) be aware the ESR MoBo is back and only limited cases will accommodate it.
PeterG
Not sure to understand everything here.
Seems in your first link you can get a complete setup with 12 cores at about 2700 bucks
Seems in your first link you can get a complete setup with 12 cores at about 2700 bucks
The first link is for Mobo, 2x 5650 CPU's, RAM and cooler. The subsequent two links are for complete PC's. They are not cheap but as I commented, pricing just for components I could not match the pre-built machine costs and of course that price includes O/C'd warranty that would be lost immediately the CPU's were overclocked if you did it yourself.
The last two links give a guide as to what components a supplier uses to build a machine.
PeterG
The last two links give a guide as to what components a supplier uses to build a machine.
PeterG
![[us]](/mwf/v12/flags/us.png "United States")
do we pay the VAT if we were to buy fom them in europe?
> do we pay the VAT if we were to buy from them in europe?
You would have to check with the supplier. Although there may be VAT saving, it is likely there would be U.S. Import and State taxes to pay if they apply. I am not familiar with U.S. duties. Buying from other European States would attract VAT, possibly at the National rate rather than UK rate if it is different.
I have not checked their O/C systems, but from a European pricing perspective this U.S. supplier always looked to have attractive prices but it would attract EEC duty when imported to European Countries.
http://www.titanuscomputers.com/
PeterG
As you said, normally you wouldn't pay the europe VAT but your country vat...
For me, buying from Switzerland, where the vat is at 8%, I save an other about 12% compared to the local supplier VAT at close to 20%.
For me, buying from Switzerland, where the vat is at 8%, I save an other about 12% compared to the local supplier VAT at close to 20%.
Hi Peter,
That is the same machine i purchased a few months ago.
That is the same machine i purchased a few months ago.
Hi Mark, powerful machine. How are you finding its performance with the various engines?
regards,
Peter
regards,
Peter
Thanks, Rybka is far better then H2 at long time controls, Critter is also very good and i just love Komodo cannot wait for Multi core version.
> Thanks, Rybka is far better then H2 at long time controls,
You have to be kidding. How many games have you run to get to this conclusion, and what was the time control?
> Critter is also very good
Critter = Houdini IMO with a few tweaks.
> i just love Komodo cannot wait for Multi core version.
We both agree on that!
I think what he is saying is on the corr chess level.
![[mx]](/mwf/v12/flags/mx.png "Mexico")
> Critter = Houdini IMO with a few tweaks.
To me they seem very different, mainly, in the way Houdini and Rybka were too similar in their move choices, while Critter remains unique in its move choices and evaluations. For real life positions, anyway.
> To me they seem very different, mainly, in the way Houdini and Rybka were too similar in their move choices, while Critter remains unique in its move choices and evaluations. For real life positions, anyway.
OK, I'll look at this again. It's quite easy to have your point of view tainted by people's comments, so I'll look at Critter afresh!
ah ok, Corr may be different. Seems a bit subjective though.
Sorry, yes i meant at Corr times.
Just out of interest, how did you reach that conclusion? I'm not saying it's wrong but did you get there by gut feel or was there something else which convinced you?
It seems Mark and me have the same feelings, I actually put my thoughts on this as objectively as I could here.
But it turns out I did in Spanish, so here's a translation:
"[...] Comparing [Rybka] with Houdini, I can tell you the following points:
-Both choose plans that are very similar, so it's not very effective to use them together on the same game, due to idea redundancy.
-Houdini has its evaluations more flattened (closer to 0) so it's more difficult to find attractive positions that you want to reach, than with Rybka
-Because of the flattening, Houdini tends to give the same score, or scores very similar, to several moves after analyzing them interactively, which isn't useful at all since its analysis becomes a list of moves. Rybka instead, has more preference for specific moves and plans to follow.
-Houdini is more pessimistic about positions that should have a higher score, I have reached positions that are won by black that Houdini gives a very low -0.60 while Rybka gives them a more correct -1.50.
-Very bad evaluation of some positions in the middle game, Rybka has this problem on endgames, where it may show 2.00 scores to drawn positions, but this problem is more serious in the middle game in which depending on Houdini can be dangerous."
For short, when analyzing with several engines, it's important that an engine favors a plan, the best one, and forces it into the other engines after cross checking (if it can't, it means the plan doesn't work, and the engine refuting it has a better understanding of the position). If Houdini doesn't have a favorite plan because it gives flat scores to most of them, then it's not very useful.
These problems that plague Houdini don't plague Critter so I don't think it's fair to say it's the same engine with tweaks.
But it turns out I did in Spanish
, so here's a translation:"[...] Comparing [Rybka] with Houdini, I can tell you the following points:
-Both choose plans that are very similar, so it's not very effective to use them together on the same game, due to idea redundancy.
-Houdini has its evaluations more flattened (closer to 0) so it's more difficult to find attractive positions that you want to reach, than with Rybka
-Because of the flattening, Houdini tends to give the same score, or scores very similar, to several moves after analyzing them interactively, which isn't useful at all since its analysis becomes a list of moves. Rybka instead, has more preference for specific moves and plans to follow.
-Houdini is more pessimistic about positions that should have a higher score, I have reached positions that are won by black that Houdini gives a very low -0.60 while Rybka gives them a more correct -1.50.
-Very bad evaluation of some positions in the middle game, Rybka has this problem on endgames, where it may show 2.00 scores to drawn positions, but this problem is more serious in the middle game in which depending on Houdini can be dangerous."
For short, when analyzing with several engines, it's important that an engine favors a plan, the best one, and forces it into the other engines after cross checking (if it can't, it means the plan doesn't work, and the engine refuting it has a better understanding of the position). If Houdini doesn't have a favorite plan because it gives flat scores to most of them, then it's not very useful.
These problems that plague Houdini don't plague Critter so I don't think it's fair to say it's the same engine with tweaks.
> -Houdini has its evaluations more flattened (closer to 0) so it's more difficult to find attractive positions that you want to reach, than with Rybka
Most of your objections seem to come from the "flat" Houdini evaluation.
Why don't you simply multiply Houdini's evals by 2, if that makes them more convenient for you?
> -Very bad evaluation of some positions in the middle game, Rybka has this problem on endgames, where it may show 2.00 scores to drawn positions, but this problem is more serious in the middle game in which depending on Houdini can be dangerous."
Houdini 2 is with all measurable time controls (up to 40/40 with 4 CPU) about 50 to 70 Elo stronger than Rybka 4.
Surely there are some positions where Rybka's handling is better, but the strenght difference indicates that in a majority of positions Houdini will pick the better move.
> Why don't you simply multiply Houdini's evals by 2, if that makes them more convenient for you?
Because then its evaluation is wrong in quiet positions. Suppose that now all engines give the position a 0.20 eval, which is accurate, if I multiply by 2, Houdini would give a 0.40 to the position, which would be too high. A different engine may see the difference between this 0.20 position and another of 0.40 where white has twice the advantage, while Houdini may give 0.20 to the first and 0.25 to the second. Doubling gives 0.40 and 0.50, which means the score for both positions is too high, while their difference isn't notable enough. Instead of trying to correct the flat scores, one is better just using the other engine.
> Houdini 2 is with all measurable time controls
I've been talking about Houdini 1.5, and from the reports I've heard, other than its learning features, people using it for correspondence games don't notice a high difference.
>Houdini will pick the better move.
What move the engine picks loses relevance on interactive analysis of correspondence games, because in critical positions the main moves of the engines are usually refuted (and in non-critical positions the main moves are evident, e.g. on recaptures you don't need the best engine to point them out), and what matters is the unrefuted moves, what is left when you refute the main ideas, an engine showing high contrast on the top moves left that one is analyzing is more useful than one that shows different tones of grey for the moves, even if the evaluation is more accurate (after all, all these moves are 0.00 if chess is solved, but against an unperfect opponent it matters what moves make the chance of the opponent making inaccuracies higher, and this requires contrast.)
Anyway, Barnard is a Houdini supporter and believes both that Houdini is the best engine around and that only one engine is necessary for analysis of correspondence games, we're playing a match to test it but unfortunately it may be over when Houdini 4 is around (but that would give him the edge as I'd not be able to use any Houdini even if it's top engine and improved a lot by then.)
Basically you're saying that it's better to have higher relative score differences, even if the differences don't accurately reflect the prediction of the outcome of the game.
That's contrary to Houdini's approach, I want to show evaluations that are as close as possible related to the probability distribution of the outcome of the game.
That's contrary to Houdini's approach, I want to show evaluations that are as close as possible related to the probability distribution of the outcome of the game.
> That's contrary to Houdini's approach, I want to show evaluations that are as close as possible related to the probability distribution of the outcome of the game.
So you would want Houdini to show a 0.00 score for all opening moves of white from the opening position, since with perfect play the best white can achieve is a draw? Because, that's not useful, as I said, Houdini becomes a move list and I already know what are the playable moves...
Otherwise, Houdini is doing a poor job of seeing the difference between some positions at the tails of the variation, expecting the attacking side to do as well on them, while another engine may have a preference.
What is the user meant to do when after interacting with the position and backtracking scores Houdini gives 0.30 score to 3 different moves? Pounce on them until they're all 0.29? Or switch to an engine that has a clear preference of a move and that can show why it's best after crosschecking the positions? (and with Houdini it'd be easy, it all needs to do is make it show a 0.31 score for it). I've found the latter much more effective, specially when I learn about a plan of a side that can be played in different position (this may not be something that happens on a root position, but that happens on a critical position down the line).
So don't just look at the main move that Houdini suggests when put into Infinite Analysis, look also at what is the second favorite move and its score relative to the first when you exclude via searchmoves, and to the third and so on, those are the moves likely to become main after you refute the main choices, and it's not useful when Houdini ends giving them very similar scores, even if they don't transpose, specially when other engines would also point to them, but with better fine-graining.
>> That's contrary to Houdini's approach, I want to show evaluations that are as close as possible related to the probability distribution of the outcome of the game.
> So you would want Houdini to show a 0.00 score for all opening moves of white from the opening position, since with perfect play the best white can achieve is a draw? Because, that's not useful, as I said, Houdini becomes a move list and I already know what are the playable moves...
Are you willfully trying to misunderstand what I'm saying?
White scores about 55% in the opening position.
This is reflected by Houdini's evaluation for the better moves like d4, e4 and Nf3 - about +0.20.
Other moves will give no advantage, for example after a3 Houdini evaluations are close to 0.00 meaning that the expected score for White is about 50%.
It is nonsense to claim that other engines have "better fine-graining" in their evaluations, if they did they would actually beat Houdini. Other engines may show "larger scores" or "larger differences", but these don't offer any improved accuracy - as Houdini's stronger rating demonstrates.
+1
> It is nonsense to claim that other engines have "better fine-graining" in their evaluations, if they did they would actually beat Houdini.
This is not about games, it's about analysis. You keep claiming that Houdini makes the best move choices and that's why it beats other engines, I do not dispute this, but for analysis of chess positions what move the engine chooses is not as important as how much fine-graining it puts when analyzing several moves. The user wants to know the difference between the top playable moves and which one the engine thinks is the best. This fine-graining doesn't appear in games, because engines only check their favorite move and don't care about how worse are the rest, while in analysis exactly how worse are the other ones is very important.
Scoring 55% on the position doesn't say the full story, perhaps the chances that the opponent draws the game or that he beats you are equal, perhaps the chance that the game is drawn are much higher but the chances that you lose the game are basically non-existant, perhaps the chances that the game is draw are very low, and the position is dangerous and double edged. If the position has 3 moves that would score 55% under these conditions, it's not useful to the user if they have the same score, Rybka Dynamic would give a boost to the third position, while Critter may like better the first one, and up to their horizon one can convince the other that their plan is better. With Houdini there's nothing to do, it's basically saying "all 3 moves are equal, so just throw a dice or something."
You're making huge generalizations here, I don't think they're correct but I guess we can agree to disagree :).
> That's contrary to Houdini's approach, I want to show evaluations that are as close as possible related to the probability distribution of the outcome of the game.
Is it therefore correct to say that +1.00 (for example) represents a certain probability of winning the game? What is it, 70%? Which evaluation = 100%?
Secondly, do you think Houdini 3 will continue with this type of evaluation? What you're doing makes a lot of sense, but it takes quite a bit of mental adjustment when working with other engines. What would be really great would be 2 evaluation windows: one showing a traditional score in pawns, the other showing the probability of winning.
> Is it therefore correct to say that +1.00 (for example) represents a certain probability of winning the game? What is it, 70%? Which evaluation = 100%?
Correct. I don't recall the numbers exactly (I did the calibration for Houdini 1.0), but +1.0 should be about 85% score.
Above +2.50 Houdini should score at least 99%.
> Secondly, do you think Houdini 3 will continue with this type of evaluation? What you're doing makes a lot of sense, but it takes quite a bit of mental adjustment when working with other engines. What would be really great would be 2 evaluation windows: one showing a traditional score in pawns, the other showing the probability of winning.
Yes, Houdini 3 will continue its tradition of relatively low scoring.
If I find the time to remake the calibration with Houdini 3 I will publish the correlation between evaluation and expected game result.
85% for +1.00 at fast time controls might well be right, but it is surely wrong at correspondence times control. +1.00 from Houdini at correspondence yields way higher than 85%. I have yet to reach a game position where +1.00 did not lead to a decisive outcome and can only think of one analysis line with a Houdini score of over +1 that led to a draw (rook + a-pawn + c-pawn vs rook, which would be reoslved in H2 by use of 6 man TBs).
In fact in I can only think of one correspondence game of mine in which a game psoition with +0.7 from Houdini led to what may have been a drawn position, though I managed to win that too, and I am not sure whether the position was defensible with perfect play.
Actually it is one of the things I like about Houdiini. When it shows a big score, you know the fat lady can start to sing. With other engines it is less clear.
In fact in I can only think of one correspondence game of mine in which a game psoition with +0.7 from Houdini led to what may have been a drawn position, though I managed to win that too, and I am not sure whether the position was defensible with perfect play.
Actually it is one of the things I like about Houdiini. When it shows a big score, you know the fat lady can start to sing. With other engines it is less clear.
True, the percentages may vary with time control (and my calibration obviously couldn't be done at correspondence TC).
At faster TC I have seen quite a lot of games with about +1.00 advantage for Houdini that finally ended in a draw, I don't expect Houdini to score more than 90% at this level.
It's also quite possible that the evolution of Houdini has changed the percentages, so I really should redo the calibration with Houdini 3.
At faster TC I have seen quite a lot of games with about +1.00 advantage for Houdini that finally ended in a draw, I don't expect Houdini to score more than 90% at this level.
It's also quite possible that the evolution of Houdini has changed the percentages, so I really should redo the calibration with Houdini 3.
Are u working in the Houdini team ?
He is the Houdini team! :)
![[pt]](/mwf/v12/flags/pt.png "Portugal")
I enjoy this idea!
I guess that a centipawn is not a centipawn anymore!
I guess that a centipawn is not a centipawn anymore!
>Barnard is a Houdini supporter and believes both that Houdini is the best engine around and that only one engine is necessary for analysis of correspondence games
not right at 100 % Uly:i told you 1 engine can be enough to play correspondence chess,and with that 1 only engine,you can achieve good results;that is different from saying ''only one engine is necessary for correspondence chess''
and yes,Houdini is the stronger chess engine,that is why im playing that match with you,me only using Houdini,and you using all the engines that you want,except Houdini (and all derived,etc...)
>but that would give him the edge as I'd not be able to use any Houdini even if it's top engine and improved a lot by then.)
well,from your own words,rybka is the better engine to play correspondence chess,and vas is ''working'' in rybka 5 (or at least he tells that to have people expectant),so you must not to be in any real disadvantage,and you also can use Komodo,that is being improved a lot,and that engine,is really developed,not like rybka ans its forgotten promisses
appart from that,my computer is start crashing and sometimes freezing,showing me a 'blue screen' about i dont know something the 'hardware'?¿,so i think my laptop has its life near to finish,and if it dies,i wont be able to buy another one in a time,so we will must hold our match until i will have money to buy another one,and that wont be in a shor time,since im unemployed (i hope it doesnt die before WBCCC 2 finish,or sadly,i will must withdraw from the tournament)
>appart from that,my computer is start crashing and sometimes freezing,showing me a 'blue screen' about i dont know something the 'hardware'?
It sounds like you need to clean it with an air blower, computers start doing that when they overheat.
>,so i think my laptop has its life near to finish
No, you're probably only need to open it and clean the inside.
where can i buy or have an air blower to clean the laptop?because i remember that i saw something simelar to a 'can' (spray) that only blows air,but i dont know where it is sell
and about being overheat,maybe you are right,because sometimes the cooler i dont listen it,and the laptop is hot,and is when usually crash showing the blue screen...
and about being overheat,maybe you are right,because sometimes the cooler i dont listen it,and the laptop is hot,and is when usually crash showing the blue screen...
Any hardware store will have something like that.
where i bought the laptop,they havent,is a small shop,and at my city,there arent big hardware store Jimmy
You don't need a big hardware store for that, anyway, for short:
El problema es que la Laptop está polvosa, abrela y limpiale el polvo, y ya está.
El problema es que la Laptop está polvosa, abrela y limpiale el polvo, y ya está.
ah...mañana la abro y le limpio el polvo del ventilador...
pero como se rompa el ventilador y haya que comprar otro,o como se rompa algo y se caliente o lo que sea,voy apañado,porque ahora no puedo comprar una laptop nueva...gracias
pero como se rompa el ventilador y haya que comprar otro,o como se rompa algo y se caliente o lo que sea,voy apañado,porque ahora no puedo comprar una laptop nueva...gracias
Or you can do it the other way with a powerful vacuum cleaner, which has always worked well for me.
Alos, someone on here said they use a leaf blower.
By the way, you can check your temps by installing CoreTemp which is freeware.
Alos, someone on here said they use a leaf blower.
By the way, you can check your temps by installing CoreTemp which is freeware.
thanks,i will try
>Surely there are some positions where Rybka's handling is better, but the strength difference indicates that in a majority of positions Houdini will pick the better move.
This might be true but does not automatically follow. If we go back even a few years, GMs would gradually outplay engines but rarely win becuase at critical points the engine would handle the tactics better. With Houdini's undoubtably superior search, the same could easily be happennig in its battles with other engines.... perhaps very often it plays slightly worse moves than other engines but at critical points finds something very important that turns the game around that other engines miss.
I personally believe (but certainly cannot prove) that there is an element of truth in this.
My main conclusions after using R41, H2, and KOM4 in IDeA on correspondence games, have some similarities with yours:
1) The flattened H2 scores make them almost useless for IDeA used interactively;
2) R4 has an unfortunate tendency to maintain scores into the endgame that are much too high (close to winning) when there are no real winning prospects--this naturally distorts the scoring of earlier moves and is largely caused by excessive weight given by R4 to material advantage.
I have no firm conclusions yet on KOM4 and would be interested in getting feedback on those who have used it in IDeA.
1) The flattened H2 scores make them almost useless for IDeA used interactively;
2) R4 has an unfortunate tendency to maintain scores into the endgame that are much too high (close to winning) when there are no real winning prospects--this naturally distorts the scoring of earlier moves and is largely caused by excessive weight given by R4 to material advantage.
I have no firm conclusions yet on KOM4 and would be interested in getting feedback on those who have used it in IDeA.
1
2
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill