Author: Slater Wold
Date: 21:48:59 09/05/02
Go up one level in this thread
On September 06, 2002 at 00:09:12, Dann Corbit wrote: >On September 05, 2002 at 23:12:00, Slater Wold wrote: >[snip] >>It *always* takes more nodes to solve a problem using more than 1 CPU, than it >>does using just 1 CPU. > >That is very surprising. I suspect that a wider search will show that this does >not always hold up. It would be pretty astonishing if it did. Of all the positions I've tested, it takes more nodes to finish a ply with more than 1 CPU. Which of course means more nodes to find a solution. If I could find a set of, say, 30 positions where it took Crafty like 2 minutes to find the solution, I could test a lot of things better. I am currently looking. (I chunked WAC. Most were solved at root or under 10 seconds.) >>It *always* takes less time to solve a problem using more than 1 CPU, than it >>does using just 1 CPU. > >This seems like it should be true. It 'tis. >>And that would be more/less what I am trying to "coorelate" here; Nodes To Solve >>vs Time To Solve. > >What does solve mean? Do the programs always have identical depth (including >fractional plies) and identical ce? If not then the searches are different and >comparision is mostly guessing. *Usually*. Sometimes they don't. About the "more nodes per ply", this is from a random position, ran on my computer for 3 minutes using 1 and 2 CPUs. 1 CPU: Ply Time Nodes 1 0.00 827 2 0.01 1417 3 0.03 3716 4 0.04 15631 5 0.12 61293 6 0.20 130008 7 0.90 683738 8 2.21 1767471 9 5.35 4461090 10 18.09 15440525 11 49.43 43087611 12 2:48 147489495 2 CPUs: Ply Time Nodes 1 0.00 827 2 0.02 1417 3 0.03 3774 4 0.05 16421 5 0.10 67567 6 0.16 140948 7 0.61 739812 8 1.44 1916333 9 3.19 4422005 10 16.97 24181765 11 36.21 52531821 12 1:44 152901077 1.72x NPS speedup
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.