Author: Terry Ripple
Date: 19:54:48 07/06/99
Go up one level in this thread
On July 06, 1999 at 14:18:31, Melvin S. Schwartz wrote:
>
>On July 06, 1999 at 00:47:17, Brett Clark wrote:
>
>>On July 05, 1999 at 21:13:47, Melvin S. Schwartz wrote:
>>
>>>
>>>Today I have received e-mail from Mark Uniacke answering my question about hash
>>>tables for Hiarcs 7.32. I would like to thank Mr. Uniacke for providing the
>>>following information in regards to the hash table question:
>>>
>>>"Please give it as much hash tables as possible. Although Hiarcs is slower than
>>>the fast searchers it maintains hash table information continually (with a
>>>suitable replacement strategy of course) so please give it as much as you can.
>>>In game play this makes a difference. In individual test positions the advantage
>>>is less."
>>>
>>>I hope this information is of value to others who have wondered about the
>>>question of hash tables.
>>>
>>>Regards,
>>>Mel
>>
>>Mr. President,
>>
>>Thanks for all of the work that you've done on the Hiarcs hash tables. I think
>>that Mr. Uniacke's statement needs to be qualified, because if you're playing a
>>5 min. blitz game with Hiarcs, I don't think that it is going to perform at
>>optimum strength if you give it 64 MB of hash tables. There has got to be a
>>correlation between the nodes per second, the average time per move, and the
>>hash table size. The longer the time control, the bigger the hash tables.
>>Hiarcs fills the hash tables at about 1/10 the rate of Fritz, so I still don't
>>see the point in giving Hiarcs huge hash tables.
>>
>>Best regards,
>>Brett
>
>Hello Brett,
>
>Have you tried e-mailing Mark Uniacke about your concerns?
>I believe he would be the only qualified individual to answer you properly.
>
>While you may appreciate the effort I have made in regards to Hiarcs hash
>tables, judging by what I see as your response has so far been the only one, I
>can only wonder if there are other people here who really care. I have made a
>great effort to not only get the hash table info, but also a strong effort to
>make ChessBase aware of the poor way Hiarcs was marketed. It appears to me that
>I may be the only one fighting this battle. Therefore, if people here don't mind
>all the confusion caused by the way Hiarcs has been marketed, why should I get
>involved with anything anymore?
>
>Where I come from, if a product is marketed in such a manner as Hiarcs 7.32,
>people would be screaming their heads off at the company for doing that. How are
>you going to get a company to improve not only their product, but also the way
>it's marketed if you don't voice your concern?
>
>I am totally disillusioned by what I see here on this site. I have seen so many
>people running around with their tail between their legs trying to find out this
>or that and wasting precious time while the company responsible for all comes
>out smelling like a rose.
>
>Does anybody besides you appreciate my effort in getting a response from Mr.
>Uniacke and posting it here...I wonder.
>Is the response from Mr. Uniacke satisfactory in regards to what you ask -
>probably not. However, I understand what he is saying in regards to a game when
>more is better. What impact does this have on Blitz chess I do not know..
>
>I would suggest you explain your concerns regarding hash tables to Mark Uniacke.
>I am very sure he will answer your question.
>
>Regards,
>Mel
---------
Hi Mel & Brett,
I`ll say it again, the Hash formula i posted a week or so ago is working
fine, and takes into account what Brett is talking about! I took this formula
and went one step further to fine tune it. You need to do each time control
because they benifit by different settings of the Hash size and you can prove
this by taking a favorite time control and starting low on the hash size and do
a Short Hi-Mark Test to find out the KN/S and the Hi-Mark. Now increase hash
size once again by aprox. 5 MB or smaller increments if shorter time
controls,then do another Short Hi-Mark test. Proceed until you start to see a
drop off in your tests and then use even smaller increments of hash size
adjustments to get the highest KN/S & Hi-Mark.
I said to you Mel that when you use only 1 or 2 MB of Hash size that you will
get your highest reading of a Hi-Mark than ever if any Time control, that is
still true but i discovered of course that this was some how faulse, as it`s not
the "Same" Answer that the tests normally show because the KN/S was so fast that
it came up with the wrong answer! How about that!! I keep telling myself if i`am
still having fun doing this! Your right about one thing Mel, us consumers
shouldn`t need to do this and i feel perhaps that i should be getting paid for
this!! Ha Ha! that will never happen! I feel like one of the Testers!
Formula: 2 x KN/S x Seconds per move. Example: I get a 26KN/S with a Short-
Test and i want to play a 40/2hr time control= 3 min. or 180 seconds per move,
so this is 2 x 26 = 52 x 180 = 9360 KB or rounded off to 9 MB to 10 MB for this
time control. Now to go another step further to fine tune this, then you need to
figure out the average amount of Opening Book Moves! I left Hiarc`s7.32 play
Hiarcs6 in a 100 game match at 5 min. Blitz and got an average of 15 moves of
opening Book Depth, as both were using the Hiarcs7.32 Book. So now you subtract
15 from 40 = 25 you put into the formula in place of the 40. This now gives you
an average of 4.8 min. or rounded off to 5 min. or 300 seconds per move. Now
going back to formula: 2 x 26 = 52 x 300 = 15600 KB or rounded off to 16 MB.
This was very close to what i now use which is 18.432 MB for 40/2hr control.
I fine- tuned this once again by like i stated in the begining of this message
by slowly increasing the amount of Hash size and performing more Short Tests
untill i got the maximum KN/S and max. Hi-Mark possible and still come up with
the same answer in the test.
Don`t forget that you need to do this for "each" time control because the
hash size will be different in most cases for maximum results! And i know this
for a fact because i tried enough different Hash sizes and many hour of tests to
come to this conclusion.
The Hash sizes are much less than Fritz but don`t forget that Fritz fills
the Hash tables 10 times faster! Good Luck and i hope this helps as much as it
did for me.
Best Regards,
Terry Ripple
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.