kN/s and does it matter?

[Michael Blake]

I understand that Hiarcs is a very good program so am a bit perplexed after comparing the programs output information against some other engines. I would be very grateful if anyone can explain what it is that I am looking at here - I tested 6 programs against White's 23rd move in the game Carlsen v Anand, Corus 2008. Carlsen played 23. Qh4 and was awarded a '!' by Golumbev to which Anand replied Qxc2 and also got a '!' for that one. So the chess engines were looking at 23. Qh4 with Black to play.

After 5 minutes reflection time all engines found Qxc2, indeed this was chosen within seconds by all the engines and all evaluated the position between 0.00 (Hiarcs 13.2MP and Houdini 1.5aw32) to -0.42 (Comet B68). Now where I got confused was in the different reported k/Ns for each engine, Hiarcs looks very slow but is this figure telling me something different?

kN/s are given at 1 minute and then 5 minutes

Comet B68 711kN/s then 750kN/s
Crafty 20.14w32 2780kN/s and then 2894kN/s
Fritz 11SE 1507kN/s and then 1517kN/s
Deep Hiarcs 12 503kN/s then 497kN/s
Hiarcs 13.2MP 469kN/s and then 470kN/s

Why is Hiarcs seemingly so much slower than the other older engines and what does this mean? In the 5 minutes given to each engine, Hiarcs 13.2MP reached depth 19/44 whilst Houdini 1.5 reached 23/53 and Fritz 11 reached 20. This meant that Hiarcs was apparently evaluating much slower yet reached a decent position depth nevertheless.

All of the above was run on a basic dual core Windows 7 laptop.
Look forward to some thoughts,
Mickey

[RadioSmall]

I understand that Hiarcs is a very good program so am a bit perplexed after comparing the programs output information against some other engines. I would be very grateful if anyone can explain what it is that I am looking at here - I tested 6 programs against White's 23rd move in the game Carlsen v Anand, Corus 2008. Carlsen played 23. Qh4 and was awarded a '!' by Golumbev to which Anand replied Qxc2 and also got a '!' for that one. So the chess engines were looking at 23. Qh4 with Black to play.

After 5 minutes reflection time all engines found Qxc2, indeed this was chosen within seconds by all the engines and all evaluated the position between 0.00 (Hiarcs 13.2MP and Houdini 1.5aw32) to -0.42 (Comet B68). Now where I got confused was in the different reported k/Ns for each engine, Hiarcs looks very slow but is this figure telling me something different?

kN/s are given at 1 minute and then 5 minutes

Comet B68 711kN/s then 750kN/s
Crafty 20.14w32 2780kN/s and then 2894kN/s
Fritz 11SE 1507kN/s and then 1517kN/s
Deep Hiarcs 12 503kN/s then 497kN/s
Hiarcs 13.2MP 469kN/s and then 470kN/s

Why is Hiarcs seemingly so much slower than the other older engines and what does this mean? In the 5 minutes given to each engine, Hiarcs 13.2MP reached depth 19/44 whilst Houdini 1.5 reached 23/53 and Fritz 11 reached 20. This meant that Hiarcs was apparently evaluating much slower yet reached a decent position depth nevertheless.

All of the above was run on a basic dual core Windows 7 laptop.
Look forward to some thoughts,
Mickey

This is simple:a low Nodes per second means that the program is processing knowledge ,But Chessmaster , Gandalf , Mchess Pro have even lower positions per second than Hiarcs. Earlier versions of Fritz also have higher nodes per second than later ones because much knowledge has been added to later versions . Today in Computer Chess depth and nodes per second are not as important as they used to be.For example Hiarcs on a mere iPhone reaching lessor Depth and calculating much less nodes per second than Fritz 6 , can and has in fact beaten Fritz 6 in a match.( see previous posts in this forum concerning such a Match).A low position per second generally means that the program has a lot of Positional Knowledge. This is a simple explanation of course I could get more technical than this........

[Michael Blake]

Hi,
Thanks for that reply. I am very inexperienced with chess engines despite having bought a few in the past to play against. Whilst waiting for the new Hiarcs GUI to come out for the Mac I dragged out my old Windows laptop and tried those engines within ChessBase. I play a lot of Correspondence Chess and can't win a game these days so thought perhaps I need to at least 'blunder check' my moves. That posed me two problems, one I don't really understand what blunder check means, and (2) I don't know how to read a chess engines output effectively.

So, please bear with me if this is a silly question - Are you saying that because Hiarcs has greater chess knowledge built into its programming, it therefore evaluates positions slower? If so, should I interpret the higher kN/s of say Fritz to mean it is less reliable?

Thanks,
Mickey

[RadioSmall]

Hi,
Thanks for that reply. I am very inexperienced with chess engines despite having bought a few in the past to play against. Whilst waiting for the new Hiarcs GUI to come out for the Mac I dragged out my old Windows laptop and tried those engines within ChessBase. I play a lot of Correspondence Chess and can't win a game these days so thought perhaps I need to at least 'blunder check' my moves. That posed me two problems, one I don't really understand what blunder check means, and (2) I don't know how to read a chess engines output effectively.

So, please bear with me if this is a silly question - Are you saying that because Hiarcs has greater chess knowledge built into its programming, it therefore evaluates positions slower? If so, should I interpret the higher kN/s of say Fritz to mean it is less reliable?

Thanks,
Mickey

Hi
Yes indeed in the Case of Hiarcs and other slow searchers it is the chess knowledge that is the cause of slower evaluation of positions.For the great majority of Human games the higher kn/s of fast searchers like Fritz will not result in less reliability , because one does not need a knowledge based program to find the mistakes in these games. However at the highest level of Human Correspondence play say at the correspondence IM or GM level , a knowledge based program would be a much better tool to use to find mistakes in the case of a loss by a very strong correspondence player . The mistakes in high level Human correspondence games are generally extremely subtle and the evaluation of fast searchers would not be very reliable. (only in certain positions in those games).

[Michael Blake]

Thanks very much for the helpful responses.