Author: Severi Salminen
Date: 11:53:06 11/27/00
Go up one level in this thread
On November 27, 2000 at 10:56:46, Feivel Avrum wrote: >This is probably a fairly basic question but I am not sure as to the answer. >While looking at all my various chess programs and engines (all in the Fritz 6 >GUI except the 2 Tigers and Rebel), I noticed a HUGE difference in nodes per >sec, and depth. I'm on a 333Mhz and the Kn varies from a low (in a typical >echange Gruenfeld middlegame) of 7 up to 222. Depth varies from 6 up to 19. My >question is, how does nps and depth correlate to strength? It seems the medium >fast searchers are stronger positionally while the deeper searchers excel in >tactics. But then there are the anomalies like Hiarcs that search extremely slow >but are extremely strong. Are the fast searchers better than slow in Blitz? Mind if I ask how have you come to an conclusion that "medium fast searchers are stronger positionally" and "deeper searchers excel in tactics"? First of all NPS doesn't tell you much. Every program reports NPS differently. It could include QNODES or could not. Usually it doesn't include nodes below a position stored in hash table, but if it did the NPS would go way up. In PVS for example one has to research the same nodes sometimes (so these nodes are included). It is very easy to make NPS look totally different and at the same time the strenght might be the same. The same applies to depth. One could report the initial "nominal" depth when starting an iteration. Usually this is a lower bound but the actual depth might be something very different because of qsearch and different extensions (for example). The depth is not a constant but some lines are searched deeper and others are not so reporting one depth is not very informative. It it very easy to remove knowledge a little so that DEpth and NPS go up but the strength remains the same on average. I suggest not to make _any_ conclusions about playing strength based purely on NPS and Depth on different programs. Severi
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.