Author: Robert Hyatt
Date: 07:36:39 09/06/02
Go up one level in this thread
On September 06, 2002 at 08:54:21, Vincent Diepeveen wrote:
>On September 05, 2002 at 11:07:38, Robert Hyatt wrote:
>
>It should affect nps. This is the diffrence between you and
>me. You assume too much for data instead of doing tests
>correctly.
Should I post some data to show I assume _nothing_? You see, that is
the difference between you and myself. I _don't_ assume anything. I
don't wave my arms and "proof" things. I just run the tests and let
the data fall where it may.
Here goes...
Default hash size = 3M bytes
time=22.76 cpu=95% mat=0 n=8890939 fh=92% nps=390k
Going up to hash=12M bytes
time=22.52 cpu=95% mat=0 n=8809597 fh=92% nps=391k
Next stop hash=48M
time=23.90 cpu=96% mat=0 n=9347280 fh=93% nps=391k
Now I _know_ you are never going to admit you are wrong. You are simply
going to wave your arms and explain why it _always_ gets faster with more
hash, but my test was flawed because I didn't search long enough, or I
searched two long, or I ran the test in a month that has "r" in its name,
or something else.
But for me, hash size doesn't affect nps much. In the above, .1% better from
3M to 12M and no further improvement. I went to 192M on my laptop with no
change from the 391K.
Anything to say?
BTW, for those wanting to do this test, I did the following: I am going to
run it yet again, but from the opening position this time. I simply cleared
the .craftyrc file, typed "book off", "sd=12" and "go".
hash=3m (default):
time=23.27 cpu=97% mat=0 n=6253934 fh=87% nps=268k
hash=12m:
time=23.01 cpu=99% mat=0 n=6452530 fh=87% nps=280k
hash=48m:
time=22.78 cpu=97% mat=0 n=6139314 fh=87% nps=269k
Little change. 12M was a bit faster, but also searched more nodes for
unknown reasons. 48M produced a tree slightly smaller than 3M, and
the NPS was back to within 1K.
As I said, "hash size doesn't have any significant effect on NPS."
I stand by that statement, because the evidence clearly supports it.
Twice.
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.