Author: Robert Hyatt
Date: 08:53:45 11/07/02
Go up one level in this thread
On November 07, 2002 at 04:37:29, Gian-Carlo Pascutto wrote: >On November 07, 2002 at 03:58:51, Sune Fischer wrote: > >>>If your quad 400 already goes 1.5Mnps, I will not claim to be able >>>to win against the quad 700. Crafty only does 750knps here, so that's >>>a large hardware advantage. I'm claiming to be stronger, not that I'm >>>sweeping the board with you no matter what the conditions... >> >>He has that parallel search overhead, so he would need to get about 25%(?) more >>nodes. If you get 750 knps and he gets 1.1 it's should actually be quite equal, >>right? > >Yes. But not 1.6Mnps. I've asked Robert if he can turn on the quad 400, >so we can get an as good as equal hardware match. > >>Are you still going, what was the end result? > >After 20 games crafty was at +6. I'm going to play more >as I expect it to at least even out. This is good stresstesting >for the WinBoard code and the learning anyway :) > >-- >GCP The math is pretty easy here. If you are getting 750K nps, that is a starting point, although I am not sure my NPS can be accurately compared to your NPS, but maybe it can. On the quad xeon, using a speedup average of 3.0, that means my effective NPS is 3/4 of the stated NPS, which means roughly 1.2M nps when the search is running at 1.6M nps. That is the comparison. I just ran a comparable test on the quad 400, which is now online, and it reported almost exactly 1/2 the NPS of the quad 700. Partly due to the reduced clock speed, partly due to it being a pentium II xeon rather than a pentium III xeon, and partily due to 512K L2 vs 1024K L2 cache. That means the quad 400 is running at about 600K effective NPS as it reported 800K in the position where the quad 700 reported 1600K. I'll leave that machine on for the day. It should also be kibitzing analysis so long as you don't play time controls longer than 5 3. At anything longer it will automatically not kibitz as humans have used that to cheat in the past by observing their own game in a second window to see the analysis and use it.
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.