![[de]](/mwf/v12/flags/de.png "Germany")
Hi everybody,
with my q6600 on playchess I get around 600 kn/s when playing blitz on playchess.
What confuses me is that my ply only is ~16.
From other machines I see that the ply varies from 14-20 for the same kn/s.
Hence my question: Shouldn't the ply be always the same ? (for instance 18 ply @ 600 kn/s)?
If not, what influences the ply depth? Why is there such a huge difference on different machines? Is a higher ply always better, or is it a higher kn/s??
Thanks for your help,
Michael
with my q6600 on playchess I get around 600 kn/s when playing blitz on playchess.
What confuses me is that my ply only is ~16.
From other machines I see that the ply varies from 14-20 for the same kn/s.
Hence my question: Shouldn't the ply be always the same ? (for instance 18 ply @ 600 kn/s)?
If not, what influences the ply depth? Why is there such a huge difference on different machines? Is a higher ply always better, or is it a higher kn/s??
Thanks for your help,
Michael
![[nl]](/mwf/v12/flags/nl.png "Netherlands")
Hence my question: Shouldn't the ply be always the same ? (for instance 18 ply @ 600 kn/s)?
No!
If not, what influences the ply depth?
Mainly branching factor, the number of moves the program considers worth investigating in a given position.
Why is there such a huge difference on different machines?
The big difference isn't between machines, but between different positions.
Is a higher ply always better, or is it a higher kn/s??
Both, but higher ply is more prominent.
No!
If not, what influences the ply depth?
Mainly branching factor, the number of moves the program considers worth investigating in a given position.
Why is there such a huge difference on different machines?
The big difference isn't between machines, but between different positions.
Is a higher ply always better, or is it a higher kn/s??
Both, but higher ply is more prominent.
Thanks for your explanation. Much appreciated.
But I'm still wondering if there's something else, because I played a number of games with the same opponent,
and the "numbers" in all of the games were quite similar:
him: ~20.5 ply , 440 kn/s
me : ~16,7 ply , 600 kn/s
and I find the huge difference in ply a bit disturbing, thinking I misconfigured something ? (hash, ram timings, I dunno)
But I'm still wondering if there's something else, because I played a number of games with the same opponent,
and the "numbers" in all of the games were quite similar:
him: ~20.5 ply , 440 kn/s
me : ~16,7 ply , 600 kn/s
and I find the huge difference in ply a bit disturbing, thinking I misconfigured something ? (hash, ram timings, I dunno)
600 kn/s sounds unusually high to me for a q6600 quad, it's more than I get with a q6800 at 3.47 Ghz. What speed are you running at and what version of Rybka? As for your opponent, getting 20.5 ply in blitz (in the middlegame) sounds impossibly high to me. Your depth sounds correct.
I'm running at ~3,2GHz with Rybka 2.3.2a mp 64Bit.
Hmm, 4 ply difference would mean (in the same position on the board!) that your opponents hardware was more then 16 times faster. Very unlikely indeed.
![[us]](/mwf/v12/flags/us.png "United States")
Hi Everyone,
Awesome who plays in the engine room of playchess.com get more than 600kn/s from his Q6600.
By the way he is also #1 on the computer list at an ELO of 3009. Good Job Roger!
Norm
Awesome who plays in the engine room of playchess.com get more than 600kn/s from his Q6600.
By the way he is also #1 on the computer list at an ELO of 3009. Good Job Roger!
Norm
Hi everyone.
Thanks for your support Norm " PAKman " Without your continued support, my 3009 would not have been possible. You have been a great source of inofrmation for me. I want everyone to know that I have spent many a night at " PAKmans " home. I have shared many chess and success storys with Norm, and his wife. Thanks again Norm for being a great friend and mentour.
Thanks for your support Norm " PAKman " Without your continued support, my 3009 would not have been possible. You have been a great source of inofrmation for me. I want everyone to know that I have spent many a night at " PAKmans " home. I have shared many chess and success storys with Norm, and his wife. Thanks again Norm for being a great friend and mentour.
Hi again everyone. I would like to wiegh in on the 600 KNs statment. I am the guy that Norm refers too " Awesome " I use Rybka 2.3.2a like most players on the Playchess server. There are many times I get 600 kns and more. Many players... and I mean many, have seen me post high 600`s KNS and almost 700 KNs . . . some have seen me post only 550 KNs at ply 14. I am not very eager to share my settings, other then to say its mostly defualt. I can also say this, my q6600 overclocks very nicly. High memory bandiwidth is in my veiw my explaination for the higher KNs. Norm "PAKman" and I played a few games friday night "8/24/07" we were also on the phone at the time . . he reported to me I was posting 671kns at about 18 ply in 3/0 time control. I have seen more, and I have seen less.
As far as the " ply " debate, in my experince in on-line computer chess " 11 years " does Rybka, or any other engine for that matter need to think 20 or more ply to see that if you don`t make a certain move at any given point of a game to stop a bad position, or perhaps checkmate. Also consider this, The high ply, and KNs at times are only a snap shot of the readings at the time the gui polls for the information.
If guys want to see real ply, look at the notation in the " Notaion " pane as the moves fly by . . . or, open any game and look at the notaion. Next to the move is real ply. Ignore what chessbase echos to you in the lower left.
Roger Z. " Awesome "
As far as the " ply " debate, in my experince in on-line computer chess " 11 years " does Rybka, or any other engine for that matter need to think 20 or more ply to see that if you don`t make a certain move at any given point of a game to stop a bad position, or perhaps checkmate. Also consider this, The high ply, and KNs at times are only a snap shot of the readings at the time the gui polls for the information.
If guys want to see real ply, look at the notation in the " Notaion " pane as the moves fly by . . . or, open any game and look at the notaion. Next to the move is real ply. Ignore what chessbase echos to you in the lower left.
Roger Z. " Awesome "
I`m not sure if this is the right place in this forum to post this statment, but I will post it anyways . . . . Vas, thanks for ALWAYS returning my e-mails. It means a great deal to me. I may ask some dumb questions, but in my view, the dumbest questions are the ones never asked. Nuff said.
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
I agree with the idea that high memory bandwidth helps a great deal with getting higher kN/s figures. My Q6600 regularly goes over 600, the highest I recall was 712, the lowest is usually about 560.
I have 2gb which runs at 1155mhz. I remember my old E6600 seemed to run a lot faster when I first installed it, I wish now I had done some accurate testing. I also wish I knew how to do some accurate testing, but I guess I am not patient enough to learn that stuff at my age.
I am hoping to upgrade to DDR3 ram with the new X38 chipset in the not too distant future, I have seen some that runs at 2000mhz. That might go some way to proving a point I guess, as to the possibility of ram speed forming some sort of bottleneck to performance.
I have 2gb which runs at 1155mhz. I remember my old E6600 seemed to run a lot faster when I first installed it, I wish now I had done some accurate testing. I also wish I knew how to do some accurate testing, but I guess I am not patient enough to learn that stuff at my age.
I am hoping to upgrade to DDR3 ram with the new X38 chipset in the not too distant future, I have seen some that runs at 2000mhz. That might go some way to proving a point I guess, as to the possibility of ram speed forming some sort of bottleneck to performance.
Phil,
DDR3 has higher bandwidth capability than DDR2, but this comes at the expense of greater latency. For chess engines, this is a bad trade off. Most of the memory accesses done by Rybka are 64-byte cache lines (for hash reads and writes) and the time to gain access to them is determined by latency rather than bandwidth. So I'll predict that if you are an early adapter of DDR3 (the current latencies are really horrible), you will actually see a small, but noticeable decrease in performance. But that's theory, so if you do this, let us know how it works in practice.
Regards,
Alan
DDR3 has higher bandwidth capability than DDR2, but this comes at the expense of greater latency. For chess engines, this is a bad trade off. Most of the memory accesses done by Rybka are 64-byte cache lines (for hash reads and writes) and the time to gain access to them is determined by latency rather than bandwidth. So I'll predict that if you are an early adapter of DDR3 (the current latencies are really horrible), you will actually see a small, but noticeable decrease in performance. But that's theory, so if you do this, let us know how it works in practice.
Regards,
Alan
Alan,
I certainly agree that the advantage offered by DDR3 is not yet clear.
I am however amazed at how fast the speeds are increasing and the latencies are dropping. 1600 at 7-7-7 was king of the hill for about 3 days before 1800 at the same speed appeared. It appears to me that specialist manufacturers are very keen to provide DDR3 with a usable speed increase over DDR2, obviously to get people spending on the next upgrade. It's the nature of the industry that when you have got everyone on DDR2 at something like it's limit, then it's time to drive forward.
At a certain point, very high clock speeds obviously overcome the slightly higher CAS timings. 6 clocks at 2000mhz for instance must create an overall lower latency than 4 clocks at 1000mhz.
A quote from an Anandtech review...
"Keep in mind that the actual latency in nanoseconds is what really matters, so while the number of memory cycles from DDR2-533 CL3 through DD2-667 CL4, DDR2/3-800 CL5, DDR3-1067 CL7, and DDR3-1333 CL9 increases, the actual latency in ns only ranges from 11.25ns (DDR2-533 CL3) to a maximum of 13.5ns (DDR3-1333 CL9). While CL7 may sound like a high latency, achieving that with 1333 MHz memory is actually results in a time latency of 10.5ns, and of course that's with much higher bandwidth than some of the other memory speeds."
I am hoping get into DDR3 when it can be clocked to 2000mhz or more at CAS7 or less. The way things are going, that will be around teatime this afternoon.
All the best,
Phil
I certainly agree that the advantage offered by DDR3 is not yet clear.
I am however amazed at how fast the speeds are increasing and the latencies are dropping. 1600 at 7-7-7 was king of the hill for about 3 days before 1800 at the same speed appeared. It appears to me that specialist manufacturers are very keen to provide DDR3 with a usable speed increase over DDR2, obviously to get people spending on the next upgrade. It's the nature of the industry that when you have got everyone on DDR2 at something like it's limit, then it's time to drive forward.
At a certain point, very high clock speeds obviously overcome the slightly higher CAS timings. 6 clocks at 2000mhz for instance must create an overall lower latency than 4 clocks at 1000mhz.
A quote from an Anandtech review...
"Keep in mind that the actual latency in nanoseconds is what really matters, so while the number of memory cycles from DDR2-533 CL3 through DD2-667 CL4, DDR2/3-800 CL5, DDR3-1067 CL7, and DDR3-1333 CL9 increases, the actual latency in ns only ranges from 11.25ns (DDR2-533 CL3) to a maximum of 13.5ns (DDR3-1333 CL9). While CL7 may sound like a high latency, achieving that with 1333 MHz memory is actually results in a time latency of 10.5ns, and of course that's with much higher bandwidth than some of the other memory speeds."
I am hoping get into DDR3 when it can be clocked to 2000mhz or more at CAS7 or less. The way things are going, that will be around teatime this afternoon.
All the best,
Phil
Phil,
Nobody is making any money on DDR2 memory, so no doubt the manufacturers will want to push DDR3 sooner rather than later.
Anandtech is certainly right that it's the latencey in ns that counts. With DDR2 memory, it took quite a while for latencies to approach what had been standard with DDR. I assumed this would also be the case for DDR3 too, with bandwidth surpassing DDR2 fairly quickly, but latency taking longer to equalize. Maybe that's a bad assumption. Chess engines are pretty unique in terms of requirements though. Offhand I can't think of any other applications where latency is crucial, while bandwidth is irrelevant.
It will be interesting to see how you make out...
Regards,
Alan
Nobody is making any money on DDR2 memory, so no doubt the manufacturers will want to push DDR3 sooner rather than later.
Anandtech is certainly right that it's the latencey in ns that counts. With DDR2 memory, it took quite a while for latencies to approach what had been standard with DDR. I assumed this would also be the case for DDR3 too, with bandwidth surpassing DDR2 fairly quickly, but latency taking longer to equalize. Maybe that's a bad assumption. Chess engines are pretty unique in terms of requirements though. Offhand I can't think of any other applications where latency is crucial, while bandwidth is irrelevant.
It will be interesting to see how you make out...
Regards,
Alan
![[hu]](/mwf/v12/flags/hu.png "Hungary")
If this is a systematic difference, then you're either using a different Rybka version or a much smaller hash. If you have disabled pondering, this will also cost you some depth.
Make sure that you look at several games, though. There will always be some fluctuation.
Vas
Make sure that you look at several games, though. There will always be some fluctuation.
Vas
Hi,
I also find the same anonamoly on the chessbase server ... I run 4Ghz+ on my quad and constantly find opponents with a far higher ply search, even dual cores at far lower Mhz. I tend to have the higher KNS\NPS (double or triple in fact) but have seen a 4 ply difference quite often .... It has crossed my mind that some people may? be useing a cluster machine? with the results coming through the server connected machine but really dont know. You can see some odd stats at times ie, it may on the screen show your opponent is at 18.6ply with a machine running a QX6700 2.66Ghz Quad at 3Mhz??????????????????????????????, it does not have to be a quad it is the odd Mhz speed at the end that always puzzles me.
Reagrds
I also find the same anonamoly on the chessbase server ... I run 4Ghz+ on my quad and constantly find opponents with a far higher ply search, even dual cores at far lower Mhz. I tend to have the higher KNS\NPS (double or triple in fact) but have seen a 4 ply difference quite often .... It has crossed my mind that some people may? be useing a cluster machine? with the results coming through the server connected machine but really dont know. You can see some odd stats at times ie, it may on the screen show your opponent is at 18.6ply with a machine running a QX6700 2.66Ghz Quad at 3Mhz??????????????????????????????, it does not have to be a quad it is the odd Mhz speed at the end that always puzzles me.
Reagrds
![[ca]](/mwf/v12/flags/ca.png "Canada")
3GHz probably, not MHz?
If so, the case is that they simply overclocked their CPU from the original factory clock (2.66GHz) to something faster (3GHz).
If so, the case is that they simply overclocked their CPU from the original factory clock (2.66GHz) to something faster (3GHz).
No, really 3Mhz. I have seen it also.
Hi,
Definitly 3Mhz and I have seen others ... you get used to seeing everyones cpu speeds, these odd ones are unusual that is why they stand out. Still no one has answered why if someone is producing up to twice the KNS\NPS of his opponent but is also getting up to 4 ply less (it does vary) shown in depth search ... Either the fritz interface is not giving the correct results or maybe altering the defaults engine settings to ultra slow and change the outlook setting might have an effect?
Regards
Definitly 3Mhz and I have seen others ... you get used to seeing everyones cpu speeds, these odd ones are unusual that is why they stand out. Still no one has answered why if someone is producing up to twice the KNS\NPS of his opponent but is also getting up to 4 ply less (it does vary) shown in depth search ... Either the fritz interface is not giving the correct results or maybe altering the defaults engine settings to ultra slow and change the outlook setting might have an effect?
Regards
3 MHz is shurely a software bug, i saw that frequently, too.
The kn/s and ply that your computer shows depend mainly on the first game after loading engine. These values are stored in gui (i think) and further games do alter these values only slightly.
So if you start your day with a long game, your computer will indicate a very high ply value, even if in the following games you play bullet. If you start off with a game with a long (mostly drawn) endgame, e.g. with only bishops and pawns, your computer will show a very high number of kn/s as long as you don't reload the engine.
So these values don't tell too much about your opponent's strenght.
Regards
Lukas
The kn/s and ply that your computer shows depend mainly on the first game after loading engine. These values are stored in gui (i think) and further games do alter these values only slightly.
So if you start your day with a long game, your computer will indicate a very high ply value, even if in the following games you play bullet. If you start off with a game with a long (mostly drawn) endgame, e.g. with only bishops and pawns, your computer will show a very high number of kn/s as long as you don't reload the engine.
So these values don't tell too much about your opponent's strenght.
Regards
Lukas
Hi Everyone, I am a new poster here. I read the forum on occasion. I am surprised at some who post here that there are un-informed. I thought it was no secret that the " 3mhz " display was a direct result of using the " Zapchess " gui . . . There you have it !
Well, the ZAP! GUI i've got reports cpu speed correctly - in case you don't believe it, here's a screenshot:
http://img440.imageshack.us/img440/5058/zap2pr3.jpg
http://img440.imageshack.us/img440/5058/zap2pr3.jpg
Hi,
I hope I did leave the impression that all ZAPchess gui`s had this error. I am aware that most of them report correctly. I have asked some of the guys on Playchess why it reports " 3mhz " , they all tell me its from the Zap gui. I know there are some issues with that gui in regards to 32 bit vs 64 bit.
I hope I did leave the impression that all ZAPchess gui`s had this error. I am aware that most of them report correctly. I have asked some of the guys on Playchess why it reports " 3mhz " , they all tell me its from the Zap gui. I know there are some issues with that gui in regards to 32 bit vs 64 bit.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill