Main
Date: 30 May 2008 20:46:07
From: Buller
Subject: Mephisto Master specs
Does anyone know the hardware specs of the Saitek Mephisto Master? Some
companies (e.g., Novag) publicize them, but Saitek apparently doesn't (other
than obvious things like number of LCD segments and a few other numbers).
I'm interested in processor, RAM and ROM size, hash table or not, opening
book size, and anything else. Some advertisements say 35,000 book moves,
others say 50,000. I've read that it has the same program as some other
machines (Senator, Milano Pro, ...), with one of the differences being which
ones have hash tables. I assume that the RAM size would reveal whether it
had hash tables or not.

I'm considering that unit because my Novag Star Diamond seems to be
developing a hardware problem, occasionally turning itself off when I press
"new game," and flashing the "low battery" display in the middle of a game,
even though the batteries are new and it's using AC power. Any thoughts on
how to get it fixed (a new adaptor didn't fix it)? Novag says I can send it
to Hong Kong for possible repair, but that seems a bit extreme. I wonder
what it'd be worth on eBay?

By the way, the Star Diamond has 118K RAM for hash tables. Any thoughts on
how much strength that really adds? With PC programs using a couple of
orders of magnitude more memory for hash tables, I wonder if 118K is really
adding much, other than in the rare simple endgame position where it can go
deep and see the right move.

Thanks,
Darryl






 
Date: 30 May 2008 16:15:25
From: Guest
Subject: Re: Mephisto Master specs
"Buller" <[email protected] > wrote in message
news:jIZ%j.71$Ev.4@trndny09...
> By the way, the Star Diamond has 118K RAM for hash tables. Any thoughts
> on how much strength that really adds? With PC programs using a couple of
> orders of magnitude more memory for hash tables, I wonder if 118K is
> really adding much, other than in the rare simple endgame position where
> it can go deep and see the right move.

118k can have quite an effect in end games with few pieces.

More importantly, it can allow researches (during the search itself) to
occur very quickly. The nodes it would need would have been the last few
nodes that were put into it. This means PVS or negascout or MTD(f) (or
whatever search method it uses) can do a research nearly instatly.

You could also do a IID (a search within a search) and much of the
mini-search would still be in the hash table, which would help the main
search.

So the hash table may not be so much for strength as for searching.


It's also worth pointing out that Ken Thompson's Belle (back in the early
80s) had a rating of 2200 and used only a 128k hash table. (True, that was
state of the art back then, but by todays standards, it was pretty limited
hardware. Much like what a small handheld chess computer could acomplish.)

A small 118k trans table isn't going to catch the kind of transpositions
that a 2gig table would, but it can still help the search and endgames.

Once you have trans tables at all that can do the basic stuff, doubling the
size of the trans table only has a relatively small improvement in the
program strength. Unfortunately I can't remember the numbers from various
messages I've read about it. It probably depends on the program and
wouldn't scale linearly anyway. (Meaning going from 64k to 128k is likely
to have a bigger relative effect than going from 64m to 128m.)


My guess is the memory was added for the hash tables because 128k was cheap,
it helps the search itself, and whatever normal transpositions it picks up
are just a nice bonus.


Also, as for the PC memory.... Although it does help, it also slows down
the search. Every hash table check basically causes the program to come to
a full stop for 200 to 400 or more cycles. A long time for a computer. And
it'll be happening at every node.

With a handheld, the hash check will be with much fewer cycles latency, so
even though the table isn't as big, the hash checks aren't slowing the
program down as much.






----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---


  
Date: 02 Jun 2008 07:56:55
From: Gian-Carlo Pascutto
Subject: Re: Mephisto Master specs
Guest wrote:
> "Buller" <[email protected]> wrote in message
> news:jIZ%j.71$Ev.4@trndny09...
>> By the way, the Star Diamond has 118K RAM for hash tables. Any thoughts
>> on how much strength that really adds? With PC programs using a couple of
>> orders of magnitude more memory for hash tables, I wonder if 118K is
>> really adding much, other than in the rare simple endgame position where
>> it can go deep and see the right move.
>
> 118k can have quite an effect in end games with few pieces.
>
> More importantly, it can allow researches (during the search itself) to
> occur very quickly. The nodes it would need would have been the last few
> nodes that were put into it. This means PVS or negascout or MTD(f) (or
> whatever search method it uses) can do a research nearly instatly.
>
> You could also do a IID (a search within a search) and much of the
> mini-search would still be in the hash table, which would help the main
> search.

The most important advantage wasn't mentioned: during the normal (not
internal) iterative deepening, the program will have information about
the best moves in many positions from the previous search.

--
GCP


   
Date: 02 Jun 2008 09:27:07
From: Guest
Subject: Re: Mephisto Master specs
"Gian-Carlo Pascutto" <[email protected] > wrote in message
news:[email protected]...
> Guest wrote:
>> "Buller" <[email protected]> wrote in message
>> news:jIZ%j.71$Ev.4@trndny09...
>>> By the way, the Star Diamond has 118K RAM for hash tables. Any thoughts
>>> on how much strength that really adds? With PC programs using a couple
>>> of orders of magnitude more memory for hash tables, I wonder if 118K is
>>> really adding much, other than in the rare simple endgame position where
>>> it can go deep and see the right move.
>>
>> 118k can have quite an effect in end games with few pieces.
>>
>> More importantly, it can allow researches (during the search itself) to
>> occur very quickly. The nodes it would need would have been the last few
>> nodes that were put into it. This means PVS or negascout or MTD(f) (or
>> whatever search method it uses) can do a research nearly instatly.
>>
>> You could also do a IID (a search within a search) and much of the
>> mini-search would still be in the hash table, which would help the main
>> search.
>
> The most important advantage wasn't mentioned: during the normal (not
> internal) iterative deepening, the program will have information about the
> best moves in many positions from the previous search.


That's normal transposition table stuff. It's a given.

And I did mention transpositions. Just that it'd have a bigger effect in
end games.

Because, with only 118k, it's not going to have many of them left if the
search gets more than 5 or 6 ply. (Don't know the exact level, of course.)

I wonder what the stats really are... Surely I have some papers on it
somewhere or other...





----== Posted via Pronews.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.pronews.com The #1 Newsgroup Service in the World! >100,000 Newsgroups
---= - Total Privacy via Encryption =---