SMP rating influence

[hcyrano]

the strength clearly improves with an increasing number of CPUs, at least up to 4 (I have almost no data for more than 4 CPUs).

old test on my othello program

acceleration :

1 core : 1
2 cores : 1,9
4 cores : 3,2
8 cores : 5,6

[Greg Strong]

t is therefore probably badly tuned for computers with 4 or more CPUs.

what are the possible settings?

It is likely more complicated than that. it would probably require programming modifications rather than simple tuning, because you first have to determine what the issues are before you can fix them. Not knowing them in advance makes it unlikely that code was included to address every case if properly modified by tuning parameters...

Yes. I meant "tuning" in a very broad sense of the word. My parallel search works, sort of -- it doesn't crash, and the strength clearly improves with an increasing number of CPUs, at least up to 4 (I have almost no data for more than 4 CPUs). But because I have never seen my program run on a quad, there are probably many inefficiencies which I am not aware of, and which I will not be able to address until I am able to test my program on a quad.

I don't worry much about this. By the time quads become common (and I don't think this will be very soon, as the current trend seems to be that tiny notebooks, "netbooks" and phones are becoming the most popular computing devices), I will probably have a quad myself.
Tord

Good to know. My new rewritten ChessV will be my first attempt at an SMP engine, and it's my current plan to model the parallel search algorithm after Glaurung's pretty closely as I don't really have a good understanding of paraller search at all.