Author: Robert Hyatt
Date: 21:36:37 12/10/98
Go up one level in this thread
On December 10, 1998 at 23:26:26, Mark Young wrote: >On December 10, 1998 at 20:55:34, Robert Hyatt wrote: > >>On December 09, 1998 at 08:44:32, Lanny DiBartolomeo wrote: >> >>>On December 09, 1998 at 08:35:51, Robert Hyatt wrote: >>> >>>>On December 09, 1998 at 06:36:02, Mark Young wrote: >>>> >>>>>Many of use have played over the games of Deep Blue Vs GM Kasparov , and Rebel >>>>>10 Vs GM Anand. >>>>> >>>>>My question is what do you think would be the stronger chess program, and by how >>>>>much: >>>>> >>>>>Deep Blue, or Rebel 10 (K6 450Mhz) * 1000 >>>> >>>> >>>>this question makes no sense. In the two longer games Rebel had no real chance, >>>>while deep blue won the match it played. You aren't going to see a rebel*1000 >>>>in the next hundred years, probably. Because technology is nowhere near even >>>>thinking about machines with picosecond cycle types. The technology to produce >>>>such technology doesn't exist yet... >>>> >>>>The next problem is that a special-purpose piece of hardware will *always* be >>>>orders of magnitude faster than general-purpose hardware, so that if we wait for >>>>a 1000x faster rebel, we get a 10000x faster deep blue in the process... >>> >>>I believe he's just asking (if it were possible) and both where run at the ame >>>exact settings which one would be stronger. >> >> >>then you have to define "same exact settings". IE would it be fair to play >>Hiarcs vs Rebel, where both ran at the exact same NPS? Hiarcs would probably >>win easily... but it would have a 5x hardware advantage to equalize NPS >>because of it's slower eval speed. >> >>So this question is *very* difficult to answer. If you mean equal NPS? Then >>the program with the better eval would win. That would be deep blue. If >>you mean an "equal handicapping" then I don't know how to do that at all... > >This is a good point. I used 1000 times faster for Rebel because I always hear >that the Deeper Blue hardware was about 1000 times faster then the current >micros. I don't know if this is true, or how people came to this conclusion. If >Deeper Blue's evaluation is doing 10 times of the work of current micro chess >program and people are just getting the speed difference from the NPS count. >Then the Deeper Blue hardware could be 10,000 times or more faster then current >micro computers hardware. I just don't know. > If you followed my comments about DB, you've seen me use the 10,000X number frequently. 250M nodes per second vs 250K for the typical micro, is a factor of 1,000. Factor in the "free computing" available in a hardwired special- purpose piece of hardware and 10X is probably reasonable based on numbers they have reported when discussing how complex their eval is (Murray or Hsu mentioned over 8,000 modifiable "weights" at one of their talks that someone reported on here...) >I asked this question because I want to know if the Deeper Blue was a much >better program then the current micro chess programs, or they were about the >same if we could speed say Rebel 10 up to the rate of a deeper blue. this is an old question, with most of its basis in "fiction." IE too many think that such a "program" is *only* about speed. I can't count the number of times I heard that same thing about Cray Blitz. Yes it was fast. But when you run it on a micro today (pure fortran version) it is at *least* 10x slower than the slowest micro program I know of. IE it is *not* "fast and dumb". Neither is DB... based on the 12 games I personally watched over 2 years... > >I could ask if we put Deeper Blue software on a K6-450Mhz would it be better >then the best micros running on the same hardware. But I don't know how fair >this would be to Deeper blue. I have heard you say that Crafty is made to run on >a fast pentium so you assume it will be. It may not be fair to slow down Deeper >Blue and try to compare. Would this be a correct assumption for Deeper Blue. not only that but it would be hard to do... since they are both hardware and software in a "symbiosis" of sorts. If you made them rewrite what they have so it would run on a PC, it may well get rolled over. But they would probably search only 1K nodes per second also...
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.