Author: Vincent Diepeveen
Date: 13:11:23 08/14/04
Go up one level in this thread
On August 14, 2004 at 15:20:57, Jorge Pichard wrote: >How are the hardwares compared as far as depth reached in the same amount of >time? > >Pichard If i understand well, Hydra runs this match at 2 processors which are i guess around 40Mhz (so 30% faster than the 33Mhz old ones) and get like 1.5 million nps a chip in hardware and shredder i guess is a quad opteron 2.0Ghz. So practical Shredder runs on 2 times faster hardware. However i do not know how efficiently Stefan is using the processors. I would guess that when you go over the top with forward pruning that the first processor is doing good work and the other 3 are only hurting you, because of all the transposition scores they together store in hashtables which when backtracked give very unreliable lines. So search depths say nothing when you effectively do not even see a ply far in a search line. I do not know how Hydra is searching, but i would guess it suffers from the same problems. Additionally i do not understand the rumours that he's using 2 processors now instead of 16. Those are however very reliable sources. What i assume is that his 12 out of 16 speedup claim is simply objectively not true as it doesn't compare 2 efficient implementations with each other. He compared a n ply search with last n-6 ply no hashtable at 1 cpu with 16 cpu's. Yet if you run dual you can of course use hashtable last n-3 ply too which is far more efficient (last 3 ply in hardware). So practical his speedup must be < 2.7 out of 16 which is a realistic speedup for the algorithm they use and which doesn't scale to more cpu's at all. A good start for hydra, though i expected them to start with a win. The next games will be tougher for hydra team. Stefan has of course option to use shredder 7.04 against hydra IMHO.
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.