| |
Main
Date: 25 Nov 2007 19:26:04
From: Pascal
Subject: New version of ScidPocket
|
Hi,
I released a new version of ScidPocket (chess database and playing
program) with some optimisations that makes all 6 embedded engines
compute 60% to 100% more nodes per second. My guess is that the achieved
ELO should be at least 2500 for some engines.
http://prolinux.free.fr/scid/PPC.html
Note that to take advantage of this improvement, the settings should be
those described at paragraph "speed optimisations".
Pascal
|
|
| |
Date: 27 Nov 2007 13:54:52
From: raylopez99
Subject: Re: New version of ScidPocket
|
On Nov 25, 10:26 am, Pascal <[email protected] > wrote:
> Hi,
>
> I released a new version of ScidPocket (chess database and playing
> program) with some optimisations that makes all 6 embedded engines
> compute 60% to 100% more nodes per second. My guess is that the achieved
> ELO should be at least 2500 for some engines.
>
> http://prolinux.free.fr/scid/PPC.html
>
> Note that to take advantage of this improvement, the settings should be
> those described at paragraph "speed optimisations".
>
> Pascal
I am interested in Pocket PC chess software, since I use it all the
time to train (easier than booting up a PC, that you cannot take on
the road).
So how does your program rank against Pocket Fritz 2.0? If you can
play at least 30 games against it (30 games will give less than 5%
error rate), at the same time controls and the same hardware, and post
your results, I can figure out what the Elo is (since Pocket Fritz is
on the SSDF list, or was, and I've figured out it's theoretical SSDF
Elo).
RL
|
| | |
Date: 28 Nov 2007 12:43:49
From: raylopez99
Subject: Re: New version of ScidPocket
|
On Nov 28, 9:14 am, Pascal <[email protected] > wrote:
> Indeed this link is really interesting.
> My own testing were made against Palm Hiarcs on a Tungsten E and a time
> control of 30 sec/move. Hiarcs is estimated 2400+ on this Palm, and Scid
> Pocket with Toga (on a Dell x51v, faster CPU than TE), is a bit better.
> So the ratings are all coherent.
> But the new release of Scid Pocket is more efficient and let the engines
> compute nearly twice faster. So my guess is that good engines (fruit
> family, Glaurung) should be around 2600 on such hardware (624 MHz ARM
> processor), and the gin error should not be higher than 100 points.
>
> But personaly I prefer far weaker engines (Phalanx) ;-)
>
> Pascal
Here are some notes of mine that I took about the new "Monahans PXA"
chip, and the Pocket Fritz 2.0. The vel PXA I think is the fastest
handheld chip today. The notes don't make sense except to me, but
maybe you can figure it out. The USCF ELO is an older rating pool
that is weaker by about 180 to 200 Elo points than the SSDF Elo pool,
based on historical data. I am using Pocket Fritz 2.0 on a IPAQ h2200
chip HP handheld, which I estimate runs at 30 seconds per move at 2445
USCF Elo. My rating is roughly 2200 USCF Elo. When training, I like
my opposition to be 300 Elo points or better than me.
RL
April 2007 - shows at roughly 550-1000 MIPS (see .PDF document in this
directory, "EvolutionofPCMIPScpu.pdf") that the new vel Monahans
PXA chip is about the same as a Pentium 3 Xeon chip. This would mean,
based on the SSDF list, since a Pentium 3 is half-way between a
Pentium 2 and 4, and since about 4x the Intel PXA255 (IPAQ h2200
chip), roughly 2x (60 or 75 pts for every doubling, the doubling rule)
= 120 to 150 pts. higher than this chip, or approximately at 30s/m
2445 + (120 or 150) USCF ELO = 2565 to 2585 USCF ELO, or Swedish ELO
2511 + 120 or 150 = about 2631 to 2661 SSDF ELO, roughly the same
performance as shown for a Pocket Fritz 2.0 being 2421 on the SSDF
(K6-2's are actually equivalent to second generation Pentium II's "The
K6 microprocessor was launched by AMD in 1997")
|
| | |
Date: 28 Nov 2007 01:08:58
From: raylopez99
Subject: Re: New version of ScidPocket
|
On Nov 27, 11:58 pm, "AZ" <[email protected] > wrote:
> > So how does your program rank against Pocket Fritz 2.0? If you can
> > play at least 30 games against it (30 games will give less than 5%
> > error rate), at the same time controls and the same hardware, and post
> > your results, I can figure out what the Elo is (since Pocket Fritz is
> > on the SSDF list, or was, and I've figured out it's theoretical SSDF
> > Elo).
>
> Scid pocket with the right engine plays better than PF 2.0.
> Here is a recent rating list with many programs for handhelds:http://raitpref.hotbox.ru/ct001-r.htm
>
> Best,
> Alain
Nice, but a mere 82 points difference in ELO, with unclear hardware,
is not enough for me to spend more money buying either your program
and/or new hardware (I need at least 150 points to do that--and I'm
thinking to do exactly that if my chess continues to improve--my 5
year old RISC based iPAQ uP is getting a bit old and I'm starting to
beat it too often).
I wish you well, as chess programming is the cutting edge of
programming, and perhaps some big mobile phone carrier will pick up
your game if you spice it up with fancy but fast graphics (the driving
force IMO, rather than a strong engine like chess purists want).
RL
|
| | |
Date: 28 Nov 2007 08:58:31
From: AZ
Subject: Re: New version of ScidPocket
|
> So how does your program rank against Pocket Fritz 2.0? If you can
> play at least 30 games against it (30 games will give less than 5%
> error rate), at the same time controls and the same hardware, and post
> your results, I can figure out what the Elo is (since Pocket Fritz is
> on the SSDF list, or was, and I've figured out it's theoretical SSDF
> Elo).
Scid pocket with the right engine plays better than PF 2.0.
Here is a recent rating list with many programs for handhelds:
http://raitpref.hotbox.ru/ct001-r.htm
Best,
Alain
|
| | | |
Date: 28 Nov 2007 18:14:20
From: Pascal
Subject: Re: New version of ScidPocket
|
AZ a �crit :
>> So how does your program rank against Pocket Fritz 2.0? If you can
>> play at least 30 games against it (30 games will give less than 5%
>> error rate), at the same time controls and the same hardware, and post
>> your results, I can figure out what the Elo is (since Pocket Fritz is
>> on the SSDF list, or was, and I've figured out it's theoretical SSDF
>> Elo).
>
> Scid pocket with the right engine plays better than PF 2.0.
> Here is a recent rating list with many programs for handhelds:
> http://raitpref.hotbox.ru/ct001-r.htm
>
> Best,
> Alain
Indeed this link is really interesting.
My own testing were made against Palm Hiarcs on a Tungsten E and a time
control of 30 sec/move. Hiarcs is estimated 2400+ on this Palm, and Scid
Pocket with Toga (on a Dell x51v, faster CPU than TE), is a bit better.
So the ratings are all coherent.
But the new release of Scid Pocket is more efficient and let the engines
compute nearly twice faster. So my guess is that good engines (fruit
family, Glaurung) should be around 2600 on such hardware (624 MHz ARM
processor), and the gin error should not be higher than 100 points.
But personaly I prefer far weaker engines (Phalanx) ;-)
Pascal
|
|
