Author: Vincent Diepeveen
Date: 05:17:18 12/07/99
Go up one level in this thread
On December 06, 1999 at 12:42:03, Pete R. wrote:
>IBM announced plans to develop a machine 1000 times more powerful than Deep
>Blue, to be used to model protein folding. Now the comparison to DB is a bit
>artificial since DB used custom chess chips. This new one ("Blue Gene") will
>have a million+ processors and perform a quadrillion operations per second. The
>question is, if IBM made a similar investment in a new chess monster, how much
>stronger than DB2 would it be in chess terms? A thousand fold increase would be
>what, an additional 6 ply search in the same time? What ply depth would it
>typically get to in 3 minutes? 20+? If it had a million of Hsu's latest
>processors it might even be more than a 1000 fold increase. Just musing. I
>think IBM could easily build a machine that would retain the World Champ title
>against humans indefinitely.
first of all they wouldn't search a ply deeper as it was done in hardware
processors, and those are dusty but still the same.
If they would use however the same amount of hardware processor but now in 0.18
or so and be a 1000 times faster then i guess it would search hardly
14 ply now instead of the 11 ply they searched, a depth we saw
regurarly reached in paderborn by several programs
800 million nodes a second * 180 seconds = 144 * 10^9
getting 11 ply with that in the axb5 position where my DIEP gets 19 ply
just needing a billion nodes (but with hashtables and with nullmove something
the deep blue processors don't have),
this means that their branching factor was around:
log 144*10^9 / log 11 = 10.7 so around 10.5 was the b.f.
(sorry i can't get the inverse X^Y of 144*10^9 xcalc doesn't have
an inverse of X^Y 11 at the sun, so i decreased 10.7 by 0.2)
Anyway another ply for deep blue is a nightmare to get. the deeper they
search the more nonsense the singular extensions will eat and that without
hashtables in the chessprocessors and without principal variation search
etcetera etcetera.
Vincent
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.