Author: Andreas Guettinger
Date: 14:05:50 07/19/05
Go up one level in this thread
On July 19, 2005 at 16:34:57, Anson T J wrote: >As all threads don't calculate the same, the NPS is a fair enough measurement. >I'm more interested in the NPS than time to depth. ??? Like Joachim already mentioned, this makes absolutely no sense to measure the speedup of dual vs single processor version in nps. It's like to measure the ELO in nps. Or measure the weigth of a car in miles per hour. The speedup of single vs dual processor version is given by the implementation of the multithreaded search method. In a worse case a lot of same nodes are calculated by both threads, which results in a bad speedup. This however has no effect on the NPS. It can be even the contrary. A "bad" multhithreaded search implementation can have very little communication between the threads. This may result in a high NPS rate of the "deep" version, despite of a very bad speedup, compared to a very good multi processor search routine, which needs more code for communication between the threads to ensure very little redundancy of the search tree and therefore gives a good speedup. So I beg you to reconsider the objection of Joachim. regards Andy
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.