Author: Jeremiah Penery
Date: 05:30:43 10/22/99
Go up one level in this thread
On October 21, 1999 at 17:06:18, Dave Gomboc wrote: >On October 21, 1999 at 16:05:36, Jeremiah Penery wrote: > >>On October 21, 1999 at 14:23:59, Dave Gomboc wrote: >> >>>IIRC, Amir once said that Junior extends even more aggressively than DB. >>>Unfortunately, he didn't get any quantitative evidence or examples. >> >>I think that common sense can show that this can't be true. >> >>At 200M NPS, why was DB getting 'only' 12-14 ply in 3 minutes/move? The reason >>is that all the rest of it was extensions. >> >>At 200k NPS (or whatever it is Junior gets - Even at 400k, it is still 500x >>slower than DB), how many ply can it get in 3 minutes/move? 10 ply? 11? (I >>really don't know, so someone please enlighten me. :) >> >>So, even though Junior is at least 500x slower than DB, it can reach comparable >>depths _AND_ extend more? I don't think so. > >I don't think that's the claim. Rather, the claim is that for a given level of >brute-force search, Junior will generally extend selected lines of play even >further than DB. I think it's clear that Junior's not going to search as many >plies full-width, nor do I think Amir ever suggested that. But Junior _does_ search as many plies full-width. DB really did do only 12-14 ply full-width. I know Fritz often does deeper than this in 3 min/move, Crafty can often reach these depths, (I don't really know much about depths of other programs, but I suspect most can reach at least near to this depth in 3 min/move). This is the dilemma - Where is all the rest of DB's speed going? >It's interesting, but I really don't know if it's correct or not. > >>>It's also difficult to know how much knowledge Hiarcs has, unless you happen to >>>be its author. >> >>True. However, it can be said that it most likely has more knowledge than other >>micro programs (except CS-Tal). The more knowledge a program has, the slower >>its evaluation, and so the slower the NPS, becauase it will take more CPU cycles >>to complete. Hiarcs is much slower in terms of NPS than most, and therefore >>there is more knowledge. Whether this knowledge is entirely useful always is >>certainly debatable. > >NPS arguments between multiple programs are complete bullshit. How people count >nodes is completely up to them. Hans Berliner would count nodes for B*, and >you'd think "wow, this is really impressive!" until you realized that he was >doing alpha-beta underneath those nodes! > >CST can do 700 nps, and Junior can do 1x10^6 nps, and that doesn't convince me >that CST has more knowledge than Junior, simply because I don't know what each >developer is counting as a node, never mind whether that node is visited briefly >or a great deal of time is expended on it. I was assuming a uniform count of nodes. :/ I never really considered thisit otherwise. However, I believe that it is still more likely that my statement above is at least 'mostly true'. :) >>> Sometimes I feel like I know more about DB than I do about the >>>commercials. Rebel excepted, of course, thanks to Ed's participation here... >>>which is much appreciated. >> >>I am very glad that Ed participates so much here. I believe he does as much as >>anyone to further progress in computer-chess (Rebel vs. GM, his experiments, >>such as Chess in 2010, PB-on vs. PB-off, etc.), and hope he continues to do so. >>Because of Ed's rather outgoing attitude about his program, I will buy Rebel >>before I buy any other commercial program. :) > >I'll support it too... though I am waiting for something that runs under NT. :-) As am I. :) That's why I haven't bought Rebel so far. Jeremiah
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.