Author: William Penn
Date: 09:09:33 04/30/04
Go up one level in this thread
On April 29, 2004 at 23:48:32, margolies,marc wrote: >Hi! >in the chessbase gui, there is a distinct setting for hash for the engine and >hash for tablebase look-up. unless the user suffers from some kind of alloc >problem due to paging from high hash settings, please tell me--because i don't >understand this--why a high hash setting for the engine should have particular >bearing on engine performance in the end game. >Assuming that your suggestion rings true, I might guess this could be a problem >of harddrive responsiveness from too many tablebase look-ups in PVs, but I am >hoping that you have a clearer answer for me. >Thanks!-Marc > Why? I don't know. Your guess is as good as mine. I just know that when the endgame approaches and tablebase access is high, the engine speed (kN/s) falls off to a fraction of normal. For example with Shredder 8, without tablebase access the speed in endgames may be over 500kN/s, but with tablebase access may fall below 100kN/s. That's with all 3-4-5 piece tablebases. This problem problem probably doesn't happen just with 3-4 piece tablebases, but I haven't tested it. I have experimented, and the only thing that helps to return engine speed towards a normal amount is to decrease the hash table size. For example the maximum is 784MB hash on my computer with the CB GUI8, and I have found that 192-256MB hash works much faster in endgames in general. The tablebase cache size has no significant effect in this case. This is true of all engines, although Shredder 8 seems to have more of a problem than most. I have a fairly modern & standard computer, I think: Windows XP Home, Athlon XP 2400+ processor at 2.0GHz, 1GB RAM, Maxtor 120GB 7200RPM 8MBcache hard drive. My best guess (very speculative) is that the tablebase access code is not optimized for Windows XP's pagefile, because the people who write that code are UNIX oriented and maybe don't really care what happens with Windows. So I think it's probably some kind of unnecessary churning of the pagefile. WP >On April 29, 2004 at 23:07:21, William Penn wrote: > >>On April 29, 2004 at 19:33:29, John Conte, Jr. wrote: >> >>>According to Chessbase.com, your suppose to hear clicking for a while. >>>I have a Compaq Presario AMD ATHLON XP PROCESSOR at 2.08GHz, with 80 GB HARD >>>DRIVE, and 512 MB SDRAM memory. What I do hear is a sound like a fan. I am >>>confused...please help. I realize the hash tables make a big difference in the >>>program operating at peak performance and also Chessbase.com states if you have >>>a greasy fast computer, never set the hash tables above (I forget but it was 75 >>>or 85MB. Thanks! >>> >>>jack >> >>Hash tables make a difference, but don't overestimate their importance. The >>tests I've run indicate they do NOT make a big difference - just an ordinary >>one. >> >>You don't have a greasy fast computer, but it's a good one. The advice you read >>about hash size sounds wrong for Infinite Analysis mode. It's OK to use 75-85MB >>hash for speed chess or ordinary games, but in general Infinite Analysis mode >>can use all of the hash possible. The more the better, usually. >> >>You can probably use up to about 384MB hash in Infinite Analysis mode if the >>GUI's maximum allows. If the op system is sluggish and you're running Windows >>XP, you can set the priority one notch lower via Task Manager for >>ChessProgram8.exe by rightclicking that process. That will improve the op system >>response and multitasking, and won't hurt engine speed significantly. However >>you shouldn't try to do much multitasking with this large proportion of hash >>(not much free RAM remaining) which could hang your system, or make it too slow >>to be useful. >> >>Or you can probably use 256MB hash with no problems, except possibly when the >>endgame approaches and if tablebase access becomes heavy with all 3-4-5 piece >>files. Then it may be necessary to reduce hash size somewhat to keep a >>reasonable engine speed. >> >>Let windows manage your pagefile size entirely. I've tried tweaking alternatives >>and now believe that's best. >> >>As for the sound, I suppose they're talking about the pagefile spinup delay for >>the allocated hash size. I'd forget about the clicking sound vs a fan noise >>advice. It doesn't sound useful. Instead, keep an eye on your hard drive >>activity light. It will be constantly active during the spinup, then subside. >>You may not get any significant spinup delays (or sound) with 256MB hash or >>less, but probably will with 384MB hash or more. It doesn't hurt anything in >>Infinite Analysis mode to maximize your hash size, so long as the spinup stops >>after awhile, sometimes taking several minutes. The engine speed will take off >>when the hash size has been fully allocated in the pagefile, and the hard drive >>activity light will subside. >> >>Good luck! >>WP
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.