Author: Vincent Diepeveen
Date: 06:21:45 04/19/02
Go up one level in this thread
On April 18, 2002 at 16:10:02, Sally Weltrop wrote:
>A Japanese machine records the fastest "floating point" calculation
>speed - over 35 trillion calculations per second. This is five times
>faster than the previous record holder, IBM's ASCI White system.
>
>http://www.processrequest.com/apps/redir.asp?link=XbddafaeBG
for computerchess this machine is a joke.
a) floating point is hardly used in computerchess
b) they just added up the *theoretical* maximum possible floating
point calculations a second which is nonsense for many reasons
b1) you can't use all processors at once for a single application
b2) the connection from node 1 to node 2 is not as fast as it
is when we buy a fast dual K7 and compare there the speed
from processor1 to processor2. In fact it's not even in the
same *magnitude*. So you start losing a factor 500 or so
to latency. Compare it to buying a few thousand of houses
with in each house a factory and then a bunch of slow trucks
which deliver the information that they must share that drive
between the houses. In computerchess that means that those
slow trucks are the limiting factor. How much a truck stores
is not interesting (it is for meteorologists and other sciences)
but the slowness of it is the limiting factor. Other houses can
only work on if they get the information delivered.
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.