Author: Albert Silver
Date: 10:30:40 09/17/05
Go up one level in this thread
On September 16, 2005 at 20:22:07, Uri Blass wrote:
>On September 16, 2005 at 20:11:35, Uri Blass wrote:
>
>>On September 16, 2005 at 19:54:04, Dann Corbit wrote:
>>
>>>Here are just the speed diffs and Elo values. I think that the 1.36 delta is
>>>highly skewed because the Pentium was such a big architecture jump from the 486.
>>
>>I can add that the hash tables were not the same and the faster hardware was
>>with more hash tables.
>>
>>I think that A1200 relative K6-450 is more than 1200/450 advantage and I
>>remember post of Jouni uski that found that the speed difference is more than
>>it(espacially for gandalf)
>>
>>Uri
>
>After some search here is the relevant post of Jouni uski
>
>http://chessprogramming.org/cccsearch/ccc.php?art_id=192372
"Deep Fritz 3,40
Junior 3,43
Shredder 4,73
Tiger(both) 2,67
Gandalf 5,30
In average 3,70. This gives (log2/log3,7)*84 = 45 points for speed doubling.
Previously it has been spoken about 60-70 points for doubling, so there is
diminishing returns now!
Jouni"
This would seem to confirm what I said. In any case, the chip differences would
clearly bring about speed differences. The most striking difference I have
found, though of course it is easier to guess, would be the difference between
my 2 GHz Athlon, a 2400XP+, and my PDA with an Intel XScale CPU at 624MHz. I
tested Fruit 2.1 using 12 MB hash on both platforms, to reduce the difference to
only the CPU, and came up with a performance difference of 15x more or less. It
varied a bit, but on average that is what I got. It's true that Fruit wasn't
written for the PDA CPU and that if it had been, it would no doubt be much
faster, still, I doubt we'd get anywhere close to the 3x difference the numbers
seem to indicate.
In any case, I ran Fruit on the Pocket PC test suite by Gurevich, and will
proceed to do so also for Crafty, Pocket Fritz 2, and Palm Hiarcs. Note that
according to the site, Hiarcs run on an emulator for the PPC, runs 25% slower
than its Palm CPU equivalent, but that should still leave it faster than any of
the Palm models available. It will be interesting to get some data on this.
Albert
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.