Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Which is better for Rybka?
- - By h1a8 (***) Date 2009-10-07 18:11
Which one is better in both scenarios, 1 or 2?

Scenario 1
1. A single core processor at 3.00ghz
2. A dual core processor at 1.5ghz for each core.

Scenario 2
1. A dual core processor at 3.00ghz each.
2. A quad core processor at 1.5ghz each.

Vas said something about Rybka's search being thicker and somewhat better with more cores. So I take it that the more the cores the less the efficiency but the better the search. I wonder will the better search more than make up for the speed loss.
Parent - By yanquis1972 (****) Date 2009-10-07 18:19 Edited 2009-10-07 18:22
no idea, but i would prefer the broader search myself for analysis purposes. also, it just makes sense that if u want to display multiple lines, multiple threads would be best. but i dont know if thats true; in fact, it doesnt seem to be since the different lines dont seem to get aligned to computer cores (why not? shouldnt this be an option?)

for computer-computer games, i would guess that the less cores/faster depth the better, but i don't know.

i forgot to add, the more cores u have the easier it is to free up the computer for other tasks if needed.  windows & most tasks can operate great on 1 core & obviously 3x1.5ghz is faster than 1x3ghz even w/ the scaling. just the other day i was playing call of juarez ii at maxed out graphics & i couldnt figure out why it was stuttering a bit every once in awhile. when i quit the game i remembered i had rybka analyzing w/ 3 cores. :) so more cores just makes for a much better user experience imo, even if its slightly slower when the comp is maxed out.
Parent - - By Gaмßito (****) Date 2009-10-07 19:22
I prefer option A in both scenarios.

With more cores, speed loss is biggest.

Parent - By ernest (****) Date 2009-10-08 23:11

> I prefer option A

What is option A ??? :roll:
Parent - By Uly (Gold) Date 2009-10-07 21:22
1. on both cases due to scaling inefficiencies (doubling the cores gives 1.7 speed up only.)
Parent - By grolich (***) Date 2009-10-07 22:00
For CPU intensive tasks (such as Chess engines), scaling is not the most efficient thing, so option 1. in both cases will simply search many more variations given the same amount of time than the corresponding option 2.

What you refer to as "thicker search" is simply the strategy adopted by Vas for using the multiple cores as efficiently as possible.
That's just a search strategy (which may or may not be more efficient than other strategies for dividing up the workload between the cpus/cores. I'm no expert on that).

That won't change the fact that the scaling of Chess engines is not perfect (~1.7 for each doubling. depends on the details, may be slightly better or (much) worse), so the single core computers in both cases would do better.
Parent - - By sarciness (***) Date 2009-10-09 04:03
In theory, the first options in each scdenario should be better. However, remember that not all processors are equal! 3GHz on one processor might be quicker than 3GHz on another, and the same could be true for duals or quads.

This site should help you.
Parent - - By h1a8 (***) Date 2009-10-09 05:38
I know. I was assuming that the same processors where used for the sake of argument. I was just thinking at thicker search to reach depth 17 is better than thinner search to reach depth 18.
Parent - By Uly (Gold) Date 2009-10-09 05:43
Yes, but not enough to offset the 0.6 slowdown.
Parent - By grolich (***) Date 2009-10-09 18:44
Not enough to make up for so much less search speed.
Even if the scaling was better (my opinion is that even if it was possible to reach scaling of 0.8 or even 0.9 for each doubling) , it would still be better to use the thinner but faster option (even 10% slowdown is quite big...)
Parent - By ernest (****) Date 2009-10-09 12:21


Intel Core i7 870 (2.93GHz) better than Intel Core i7-975 Extreme (3.33GHz) ???
Up Topic Rybka Support & Discussion / Rybka Discussion / Which is better for Rybka?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill