Author: Robert Hyatt
Date: 11:38:14 11/09/98
Go up one level in this thread
On November 09, 1998 at 12:12:59, Dave Gomboc wrote: >On November 08, 1998 at 14:34:00, Robert Hyatt wrote: > >>Pretty much correct. Cray blitz still exists... but is not being modified >>since Crafty was started. I started the "crafty" project after the 1994 ACM >>event in Cape May... machine time is *very* difficult to get, there is little >>room for the "unexpected" (such as a weather delay or whatever) since the >>machines are so tightly scheduled... >> >>I wearied of the process of setting up machine time every year, dozens of >>phone calls, emails, begging, borrowing, etc... >> >>The Cray's will still blow off any collection of microprocessor-based machines >>you'd care to use, but at $60,000,000 they are expensive and difficult to get >>hold of. >> >>I gave up not because the micros were catching up in speed (which they weren't, >>not even close) but because the micros are so much easier to get access to... > >In your opinion, which would be a stronger player on a speedy Cray machine: the >Cray Blitz program as it has sat unchanged for several years, including its >vector optimizations, or Crafty, compiled, possibly with a few short routines >inlined in assembly (say the equivalent of X86.s)? > >I guess what I am wondering is if it would be worthwhile for you to still try to >get access to Cray hardware for a world championship tournament to try with the >Crafty software, or if it would just be weaker than the old Cray Blitz anyway. > >If "Cray Crafty" would be a better player than the old Cray Blitz, would you >consider trying it in an important tournament? > >Dave Gomboc Probably cray blitz... because it has about 20,000 lines of assembly language, and was specifically designed around the cray architecture. It's difficult to vectorize things in crafty at present, although the bitmap stuff does work quite well with a popcnt instruction and the like ready to go on the Cray. But the vectorized move generator, attack detector, static exchange evaluator and so forth are *very* fast on the Cray. If we took fortran vs C for crafty it would be closer, but the fortran version still vectorizes as it was designed that way. the C version wasn't. Yes, I think Crafty's search with the recursive null-move and so forth would be better, but only after a lot of work to get the vectorization up to speed to really take advantage of what that machine offers... It would take several years, unfortunately...
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.