Author: Robert Hyatt
Date: 20:15:37 08/10/01
Go up one level in this thread
On August 10, 2001 at 15:43:40, Janosch Zwerensky wrote: >> For Chess, DB >>was far faster. > >So you think that it wouldn't be possible to have an ASCI white - class general >purpose supercomputer reach an effective computing power of more than two >trillion instructions per second (which would be enough to do 200 million nps >assuming 10000 operations per evaluated node) when it executed a chess program? The point is that DB did the equivalent of much more than 10K operations per node when you consider what the chess hardware did in parallel. This machine has two serious drawbacks... (1) it is not shared memory, it is a message- passing architecture. That makes a search far less efficient than what we get on SMP boxes, because a shared hash table becomes a performance bottleneck when each hash probe results in net traffic; (2) message latency is very high, which means an efficient parallel search is going to be tough. A reasonable search can be done, but there are none at present that would use this machine very efficiently. Other issues are the large number of processors makes the search quite inefficient as well...
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.