Author: Robert Hyatt
Date: 06:49:01 05/25/99
Go up one level in this thread
On May 25, 1999 at 01:56:01, Chris Anderson wrote: >On May 24, 1999 at 10:54:41, Fernando Villegas wrote: > > >> >>Hi Chris: >>There is some tiny flaw in your reasonning and it is this: DB IS mostly the >>hardware, so of course everything on it is "largely" owed to procesing power. If >>Hiarcs is putted to work in comparable hardaware, it would not be any more >>hiarcs, but a kind of DB. DFB hardware is not like a PC, a general use device, >>but ot is specificaly designed to calculet chess moves. >>fernando > > I realize this, but I asked the question because people were wondering >whether Hsu was going to go commercial with DB. Your above point seems to answer >the question and concurs with my thoughts. However is it inconceivable to run >Hiarcs or Fritz on some massive supercomputer and achieve near-DB search depths? > > Chris Anderson No. I ran Cray Blitz on the fastest machine on the planet, the Cray T932, and hit a speed of 5M nodes per second. Which is 100x slower than Hsu's monster. Cray Blitz took years to develop and make work on a vector computer. It would take at _least_ 5 years to port hiarcs and get the kind of speed we got. Just dropping in a C program (like Crafty) produces a per-cpu speed _slower_ than that of a pentium pro 200mhz machine. A commercial chess program on a supercomputer just won't happen for that reason. And even if it did, it is not in the same ballpark with DB. Not even close. As I have mentioned before, DB is the equivalent of at least 1000x faster than the best PC, because they get their hardware evaluation for free while we dedicate thousands of instructions to it.
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.