Author: Robert Hyatt
Date: 20:22:26 08/18/00
Go up one level in this thread
On August 18, 2000 at 21:46:30, Aaron Tay wrote: >On August 18, 2000 at 09:27:21, Robert Hyatt wrote: > >>Cray Blitz and crafty are _very_ similar now. > >I'm rather curious about this statement. By "similar" , what do you mean? > > CB and Crafty have similar evaluations. CB better in a few places (King safety where the attacks are cheaper to compute on a vector machine) Crafty better in others. Search extensions are similar except that CB had singular extensions while the current version of Crafty does not (there is a SE version of Crafty, but I have never been happy with it and it has been a "back burner" project). >I understand that it's not a simple matter of just coverting the code from a >mainframe to Crafty's micro mode, but did it really take that long for you , to >put all you have learnt from CB to crafty? Much of CB was in algorithmic development, more than chess related. IE a lot of work went into vectorizable move generation, attack detection, etc... once the basic ideas were chosen, it was a _lot_ of work to make those algorithms work well on the vector hardware. You are correct. Dropping a normal program (like Crafty) onto the Cray is a terrible mistake. Takes a lot of work to take advantage of all that the machine has to offer. > >Or are you saying that the initally pcs were much slower, so you couldn't put >everything you wanted from Cray Blitz to crafty untill much recently? basically, correct. At the time CB was searching 1M nodes per second on the best cray of the day, it was searching 100 nodes per second on the best PC of the day. Tactically, it couldn't survive much at 100 nodes per second. > > >>Do you think Crafty at 7M nodes per second would get rolled over very badly by >fritz at 1M? I don't. > >The opposite would be more likely.. > >But i think the assumption was that computers of today including Crafty in 2000 >is superior to Cray Blitz becuase of the advancement of new algorithims and >better understanding of computer programming in general. > That is where I have an advantage over everybody else here. I worked on Cray Blitz and know what it had inside. And I now work on Crafty and know what it has inside. The two programs are _very_ similar. If I disable singular extensions in CB, and make a one line change to the null move search to make it recursive, R=3 to 2 depending, then they would be _very_ similar. Many like to speculate that CB was only good because of the hardware. That was only _part_ of the equation. I am not doing many things differently than I did 15 years ago in CB, particularly in the evaluation and search extensions. I have (now) a much simpler quiescence search. I haven't seen any great gain or loss from doing this, other than making the code simpler and less buggy... If you think that CB would be smothered by today's micros, then you should also think that Crafty would suffer the same fate. They are _very_ close to each other, overall, in evaluation, search, extensions, etc. Except that Crafty is a 64 bit program while cray blitz was vectorized heavily. CB didn't really depend very much on 64 bit values other than for hashing and the simple bitmap attack detection we used. >Hence the claim that Fritz running at 1M would beat or at least hold Cray Blitz >because of it's superior Year 2000 "engine". That is my point. I don't believe that the 2000 "engines" are vastly different from the 1990 engines. At least mine isn't, and it is playing ok in today's world... > >But now it appears you are stating that Crafty is not a better engine than Cray >Blitz despite time passing. Correct. Maybe better in some places, but also worse in others. > >And if Crafty is very similar to Cray Blitz now, i would guess that Fritz and >the rest would be around the same or at least not so much better that it can >make up for the difference in speed..
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.