Author: A. Steen
Date: 00:28:37 11/21/05
Go up one level in this thread
On November 21, 2005 at 03:13:52, Aaron Gordon wrote: >On November 21, 2005 at 02:51:03, A. Steen wrote: > >>"Instantly" is interesting. >> >>S(KRK, Nalimov EMD) =! 7Kb, which resides in RAM cache easily. But we are >>talking about the general question of EGTB use, where tablebase sizes make RAM >>cache-residence a dream many years from reality (and even then, pre-load time >>will be an issue). So for real life use, HDD latency and seek times are an >>issue, and periods of even 10ms have to be considered. >> >>For a 4GHz CPU, even 5ms = 20,000,000 CPU cycles. With hyperthreading and >>family, that is an awful lot of processing time for one tablebase lookup. >> >>So, "instantly" is usually wrong. > >Anyone of average intellect can discern what he was talking about and would know >not to take it literally. Thanks for the fresh insult. Note what is being considered is either: * using (some) EGTBs; and/or * using a chess engine evaluation function. There is no plausible third option of just picking the move up out of a hat or out of the end of a randomising move generator. which would be time-cheap but useless. So the times taken to find the move with the help of the EGTB, or using only the evaluation algorithm, are what needs to be compared. I showed that for just one single "perfect" EGTB look-up, one might need to pipeline 20 million clock cycles worth of evaluation processing. Conclusion: EGTB lookups are very expensive. You get perfection, but it costs. Like most decisions in life, it is not always clear-cut. >This is a blatant troll, no doubts about that. I suggest you make your profile private, as you may not wish to associate yourself with these academic pronouncements when you have thought the matter through. I stand by everything I have written. The S/N here is poor indeed, and perchance the confederacy is forming. Best, A.S.
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.