Author: Uri Blass
Date: 15:50:12 11/15/02
Go up one level in this thread
On November 15, 2002 at 18:03:29, Robert Hyatt wrote: >On November 15, 2002 at 10:41:55, Uri Blass wrote: > >>On November 15, 2002 at 10:27:12, Robert Hyatt wrote: >> >>>On November 15, 2002 at 01:02:52, Uri Blass wrote: >>> >>>>On November 14, 2002 at 19:57:02, Robert Hyatt wrote: >>>> >>>>>On November 14, 2002 at 18:07:40, Uri Blass wrote: >>>>> >>>>>>On November 14, 2002 at 17:20:45, Robert Hyatt wrote: >>>>>> >>>>>>>On November 14, 2002 at 12:57:19, Rolf Tueschen wrote: >>>>>>> >>>>>>>>On November 14, 2002 at 11:26:37, Robert Hyatt wrote: >>>>>>>> >>>>>>>>>On November 14, 2002 at 03:33:48, Gian-Carlo Pascutto wrote: >>>>>>>>> >>>>>>>>>>On November 13, 2002 at 16:52:35, David Hanley wrote: >>>>>>>>>> >>>>>>>>>>>>If you play the current best program on current hardware against that >>>>>>>>>>>>combination, it's also going to blow it over. >>>>>>>>>>> >>>>>>>>>>>Against the kasparov, etc? Well, well see. But i expect that it won't >convince either camp. >>>>>>>>>> >>>>>>>>>>No. DB of then against the top of now. I suspect DB would get spanked. >>>>>>>>>> >>>>>>>>>>DB of then against the programs of then is another matter. >>>>>>>>>> >>>>>>>>>>-- >>>>>>>>>>GCP >>>>>>>>> >>>>>>>>>I'll change the metaphor a bit, but if by "spanked" you mean that DB's >>>>>>>>>fist would get beat to a bloody pulp by the faces of today's micros" then >>>>>>>>>I might agree. >>>>>>>>> >>>>>>>>>But _only_ in that metaphorical context. >>>>>>>> >>>>>>>>If it's only about metaphors, I think that computer chess is also a topic for >>>>>>>>me. I have the concrete question if you could give us a comparison from the old >>>>>>>>days. How would you compare the difference in strength between the actual >>>>>>>>commercials and DB2 in giving the names of ancient programs? Could we say, CRAY >>>>>>>>BLITZ against FRITZ 2 or what would you prefer? >>>>>>>> >>>>>>>>Rolf Tueschen >>>>>>>>I >>>>>>> >>>>>>>I am not sure what you are asking. I don't personally have a lot of experience >>>>>>>with older >>>>>>>commercials. The only experiment I ever ran caused a lot of ruckus in r.g.c >>>>>>>(prior to the >>>>>>>days of r.g.c.c) when I ran several games between a single-cpu Cray Blitz vs >>>>>>>Chess Genius >>>>>>>2 on the fastest PC of that day, which I think was a 486/66 or something >>>>>>>similar. It ended >>>>>>>like the DB single chip vs the micros ended, except that I _did_ post the games, >>>>>>>without >>>>>>>posting the name of the opponent. But someone (Chris Whittington I think) >>>>>>>figured it out >>>>>>>because it was a king safety debacle for the micro. >>>>>>> >>>>>>>All I can say about DB2 vs the micros is that it is about 200x faster. That's >>>>>>>more than enough. >>>>>>>Null-move or not. IE I wouldn't want to play a match Crafty vs >>>>>>>Crafty/no-null/200x faster, >>>>>>>myself, and that would not be a completely fair test since I know that DB did >>>>>>>some things in >>>>>>>their eval that I am not doing at present... >>>>>> >>>>>>1.Deeper blue was not 200 times faster than Crafty of today. >>>>>> >>>>>>Hsu said in reply to the question about the number of nodes that >>>>>>the 200M nodes were 200M total nodes and not effective nodes. >>>>> >>>>>So? My 1M nodes is not "effective nodes" either. Nor is the NPS for any "deep" >>>>>program... So 200x is right in the ballpark. >>>> >>>>For Deep blue the difference was clearly bigger because all of their >>>>problems(not using hash tables in the hardware and loss of speed from other >>>>factors). >>>> >>>>Uri >>> >>> >>>Not necessarily. Deep Junior doesn't hash in the last ply or two plus not in >>>the q-search. Do you think he does that because it is less efficient? Or >>>because it works _better_? >> >>Deep Junior use different algorithm >> >>I know that they did not hash and did not use killer moves in the hardware >>because they had not time and not because it worked better. >> >>Uri > > >So? The point is that it is not clear that hashing way out there is better >_anyway_. It is clear that at least killer moves are better everywhere. I think that it is also clear that hash in the last plies if you do not include the qsearch is better(otherwise I could expect programs not to do it). Uri
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.