Author: Robert Hyatt
Date: 10:48:14 01/23/00
Go up one level in this thread
On January 23, 2000 at 12:52:29, Albert Silver wrote: > >In hardware, I could be considered at best as an enlightened know-nothing, so I >won't pretend to be entering the argument here. I'm curious though. Is the SP >really no different from a PC except that it is bigger (and faster)? I mean take >the Cray for example, or any supercomputer for that matter; I would have thought >that it also involved massively different electronics permitting wondrous >things, but only if the software were specifically written to exploit these >differences. So a PC program that was simply emulated in it would be taking no >advantage of what it had to offer. This would mean that such >supercomputer-specific programs would be pretty much impossible to port over, as >the program depends on this hardware, and to be ported would have to be >literally rewritten. And even if this were successfully done would be pretty >much unusable. In other words by the time all the necessary changes were made to >make it work it would be a different program. If this were the case also for DB, >the end result wouldn't be DB, but just inspired from DB. > > Albert Silver Tom hasn't done any of this (yet). And experience is a good teacher about the difference between theory and reality. He "might" say that it would be easy to take Cray Blitz and run it on a PC. It was written in FORTRAN and there are good PC-based FORTRAN compilers. And in fact I did this. On a Cray cpu running at (back then about 80 megahertz (12.5ns clock cycle cray-1) we ran at over 20K nodes per second on a single CPU. I ran this on a 133mhz P5 and could not hit 100 nodes per second. The reason was the missing vector hardware that I depended on in CB. It just became a huge memory bottleneck problem on the PC. I _could_ have rewritten things to run faster on the PC, but that would have made it far slower on the Cray. The SP is a message passing box... no big deal. But DB uses the special-purpose hardware chips to store chess positions, make/unmake moves, etc. All of that has to move to software, along with the eval which is 100% hardware. This would take a _lot_ of time, and would be drudge work since it is already working in hardware with no hope of running within 5 orders of magnitude of the speed of the hardware when it is moved to software. My ratio was 20000:50. But I only needed vector hardware, everything _was_ software in CB. Hsu has a much more serious problem to overcome..
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.