Author: Christophe Theron
Date: 10:06:07 05/29/01
Go up one level in this thread
On May 29, 2001 at 00:36:53, Ian Osgood wrote:
>On May 27, 2001 at 19:41:04, Christophe Theron wrote:
>
>>On May 27, 2001 at 16:55:02, Ian Osgood wrote:
>>
>>>On May 25, 2001 at 04:29:36, Christophe Theron wrote:
>>>
>>>>On May 25, 2001 at 01:29:11, Mike S. wrote:
>>>>>
>>>>>I think it would be interesting to see if Palm Tiger is able to win a longer
>>>>>match against the NOVAG Sapphire II (I would expect that, but the Sapphire has
>>>>>proven superior so far against other handhelds or travel computers, according to
>>>>>info I have read from various sources).
>>>>>
>>>>>I wouldn't be surprised if the result would be quite narrow though...
>>>>
>>>>
>>>>I really don't know and I would really like to know.
>>>>
>>>>The Sapphire II is rated 2012 by the SSDF (that means 2112 in the previous
>>>>versions of the list, which is more accurate for lower ratings).
>>>>
>>>>So normally Tiger for Palm should be able to at least equal the Sapphire.
>>>
>>>So far, my Sapphire II has 2.5 - 0.5 against Palm Tiger at 5 minute games.
>>>Tiger is overclocked to 28MHz on a Palm IIIe, both have hash tables on,
>>>permanent brain disabled. Looks like the Sapphire II is out thinking Palm Tiger
>>>by about a ply (3000 nps vs 500 nps).
>>
>>
>>Tiger is not comfortable with games in 5 minutes on the Palm. That's too fast
>>for my search algorithms.
>>
>>That's something I'll work on, but until I manage to improve on this, I would
>>advice to play longer games to get more significant results.
>
>Understood.
>
>>> I plan to try some 30 minute games with
>>>permanent brain later this week. (This would be easier if Palm Tiger had some
>>>sort of serial interface so that I could adapt it to WinBoard like I have my
>>>Sapphire II.)
>>
>>
>>Unfortunately I'm not planning to work on this. Too few people would be
>>interested in this. You would need to own two Palms, or a Palm and a compatible
>>device (Sapphire?), that would attract probably no more than 10 customers. Too
>>much work for too little interest.
>
>I was hoping it might not be too much work, since you have already implemented a
>terminal-like interface. But of course it is up to you.
Having a terminal-like interface does not help here.
I would have to write a transmission protocol from scratch, and preferably one
that the Sapphire understands.
That's too much work and testing time, and I do not even have a Sapphire, so I
don't see that happening.
>And an update: at 30 minute sudden death, with hash and permanent brain, Chess
>Tiger achieved two draws against SapphireII. In both games, Tiger obtained an
>advantage in the endgame (R-N in one, Q-R in the other) but Sapphire II used
>excellent endgame technique to hold the draw. Search depths seem to have
>equalized (probably due to Tiger's modern techniques such as null-move etc.),
>with the advantage going to whoever managed to predict the last move.
>
>Next I'll do 10 and 15 minute games.
OK. Please keep us informed.
>>>>>I think the Sapphire II could act as a kind of basic performance test for Palm
>>>>>chess programs, in other words, they should beat it more or less clearly to be
>>>>>competitive IMO, in terms of strength (in terms of functionality, the Palm is
>>>>>certainly superior).
>>>>
>>>>
>>>>There are not many chess programs which are going to be able to pass the test!
>>>>:)
>>>
>>>Indeed. People forget that the SapphireII is near state of the art at its clock
>>>speed (16MHz). Lang's and Morsch's programs in Mephisto units are its only
>>>competitors.
>>
>>
>>I do not know if the Sapphire processor can do more MIPS than the Palm's
>>DragonBall. Anybody has info on this?
>>
>>I know the DragonBall takes a lot of clock cycles to process only one
>>instruction, and I remember the Sapphire has some kind of RISC processor in it.
>>Right?
>
>It has an H8 processor ("RISC Style Technology" is silkscreened on the front of
>the unit). Novag claims it is bitsliced to 32 MHz off of a 16 MHz clock. I
>don't know enough about processors to know whether this is a straight 2x speedup
>or not.
So I guess the Sapphire has more processing power than the Palm.
>>>>>A Pocket PC, or Windows CE program running at 100...200 MHz just must beat the
>>>>>Sapphire I think... at least if it is somewhat state of the art.
>>>>
>>>>
>>>>Clearly. The Sapphire should be wiped away by a decent chess program running on
>>>>StrongARM at 100 or even 200MHz.
>>>>
>>>>
>>>>
>>>> Christophe
>>>
>>>However, that program must be well written. For example, the strongest
>>>commercial program for the PocketPC, PalmChess, on a 206MHz iPAQ loses (~35%)
>>>against ChessGenius on a 28Mhz overclocked Palm. It would certainly lose
>>>against SapphireII which is stronger.
>>>
>>>Crafty on the iPAQ is stronger than the SapphireII (hash tables help quite a
>>>bit.)
>>
>>
>>The difference in processor speed between the Sapphire and the StrongARM
>>screams.
>>
>>
>>
>> Christophe
>
>But the high quality of crafty, which takes full use of the processor and
>decades of chess programming experience compared to PalmChess, is what makes the
>difference. (For giggles, maybe I'll port SCP or TSCP to the iPAQ for a
>strength comparison.)
>
>Also remember that the Sapphire II is programmed in assembly down to the metal
>and so gains in effective speed compared to a generic C program such as crafty.
Additionally, maybe the way Crafty is designed does not make it a good candidate
for 32 bits handhelds, because it uses a lot of 64 bits integers.
It is certainly a very bad choice for the Palm, which has a 16/32 bits
processor: it is able to process 32 bits integers directly, but the RAM bus in
only 16 bits wide, so it takes more cycles to read and write 32 bits integers
from/to memory, not even talking about playing with 64 bits integers!
Christophe
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.