Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Jim Ablett's build?? or Author's build??
- - By richbell (**) Date 2007-11-01 14:36 Edited 2007-11-01 14:39
I have a Dell Laptop with following config.

-------------------------
  CPU-Z version 1.41
-------------------------

Processors Map
------------------------------------------------------------------------------------

Number of processors       1
Number of threads         2

Processor 0
    -- Core 0
        -- Thread 0
    -- Core 1
        -- Thread 0

Processors Information
------------------------------------------------------------------------------------

Processor 1 (ID = 0)
Number of cores      2
Number of threads  2 (max 2)
Name              Intel Core Duo T2050
Codename        Yonah
Specification           Genuine Intel(R) CPU           T2050  @ 1.60GHz
Package             Socket 479 mPGA (platform ID = 5h)
CPUID              6.E.8
Extended CPUID     6.E
Core Stepping       C0
Technology         65 nm
Core Speed        1596.2 MHz (12.0 x 133.0 MHz)
Rated Bus speed     532.1 MHz
Stock frequency       1600 MHz
Instructions sets      MMX, SSE, SSE2, SSE3
L1 Data cache        2 x 32 KBytes, 8-way set associative, 64-byte line size
L1 Instruction cache    2 x 32 KBytes, 8-way set associative, 64-byte line size
L2 cache            2048 KBytes, 8-way set associative, 64-byte line size
FID/VID Control        yes
FID range           6.0x - 12.0x
max VID             1.263 V
Features           XD

Chipset
------------------------------------------------------------------------------

Northbridge         Intel i945PM rev. 03
Southbridge        Intel 82801GHM (ICH7-M/U) rev. A1
Graphic Interface    PCI-Express
PCI-E Link Width    x16
PCI-E Max Link Width   x16
Memory Type       DDR2
Memory Size        1024 MBytes
Memory Frequency  266.0 MHz (1:2)
CAS#              4.0
RAS# to CAS#      4
RAS# Precharge     4
Cycle Time (tRAS)    11
Bank Cycle Time (tRC)  16

My question is about latest scorpio 2.0 release. There are two different builds available.

This is by the author => http://wbec-ridderkerk.nl/html/download/scorpio/scorpio_20.rar
This works for me, and performs better.

This is JA's two special builds  => http://www.orbitfiles.com/download/id2146348244.html
(Win32 intel compiler 10 p.g.o.
Contains a universal 'all processors' build and
Intel® Core™ Duo, Quad & Xeon(R) only build.)

'All processor build' works for me. The other one supposed to work doesn't work for me i don't know why, says "not supported in this processor".

As you see my processor's instruction set supports MMX, SSE, SSE2, SSE3 and its a dual core. Why would Jim Albett's build for dual core doesn't run for me.  I run JA's all processor build fine. Also the original release available in WBEC (Non-JA release) works too.

Non-JA (WBEC version) release scores about  65 - 70%
JA-Release for all processors scores about      50 - 60%

(About 50 games so far tested)

1. Could anyone explain if JA release is superior or not and what exactly is the technical difference between author's build and JAs?
2. They say JA's win32 build is superior for all engines since the build is optimized for faster execution then why is it NOT performing well compared to author's original build in my box?

If someone clarifies which build is preferable i would highly appreciate.

Note: Sorry for a long data, i don't want to throw information back and forth and confuse, instead want to provide complete information i have the first time. :-)

  
Parent - - By Vempele (Silver) Date 2007-11-01 15:02

>This is by the author => http://wbec-ridderkerk.nl/html/download/scorpio/scorpio_20.rar
>This works for me, and performs better


How much faster is it?

>(Win32 intel compiler 10 p.g.o.
>Contains a universal 'all processors' build and
>Intel® Core™ Duo, Quad & Xeon(R) only build.)


Probably a typo. I think he meant to say Core 2.
Parent - - By richbell (**) Date 2007-11-01 15:16

> How much faster is it?


How can i measure how fast is a chess engine? I don't believe N/s is accurate in all circumstances. I am running a bullet tournament with 1/0 time control. So far Author's build performed 10%-15% better than JA's build (though it's just 50 games so far).

This shows faster build too in my opinion cause 20%-40% of bullet games are lost due to time factor, so faster build prevails eventually on same version of two engines. Let's see...

> Probably a typo. I think he meant to say Core 2.


This could be true.

Please let me know if i am missing something,  that you guys know...

Parent - - By Vempele (Silver) Date 2007-11-01 15:29

> How can i measure how fast is a chess engine?


nps, obviously. There's no other difference between the builds so it correlates perfectly with speed.

>This shows faster build too in my opinion cause 20%-40% of bullet games are lost due to time factor, so faster build prevails eventually on same version of two engines. Let's see...


If even one game is lost on time, there's something seriously wrong with your test conditions (or the engines). You should use an increment IMO.
Parent - - By richbell (**) Date 2007-11-01 15:38

> nps, obviously. There's no other difference between the builds so it correlates perfectly with speed.


Cool, i didn't know it's accurate info, cause i read once it differs with what postion your engine is, and i guessed it always differs when two engines compete even if they are same versions.

> If even one game is lost on time, there's something seriously wrong with your test conditions (or the engines). You should use an increment IMO.


Even on bullet games, is it not fine if an engine looses on time?? Won't there be poorly performing engines in speed or in other words (slow thinkers)? hmm, may be i should set an increment value. 
Parent - - By Permanent Brain (*****) Date 2007-11-03 13:20 Edited 2007-11-03 13:22
For a speed comparison by kN/s, you need to make anything else identical, position, hash size, etc. If the engine version (build version) is the only thing what's different, you get a valid comparison of course. The simplest idea would be to run them in infinite mode in the starting position with identical settings, each.

If bug-like effects like too many time losses cause a bad percentage of one version, then it isn't a speed issue (in terms of kN/s) IMO...

There are some engines which I am interested in, but after a new release I always think to myself "Ok - and now let's wait for a couple of days for the instant bug findings & fixes as usual..." :-) I am afraid Scorpio is one of them. But it's among the stronger freeware, and one of the few free "deep" engines.
Parent - By richbell (**) Date 2007-11-04 05:36
Thanks PB and Vempele. I am working on it now. I guess my time limit of 1 min without increment caused lot of timeouts. Now i have an increment of 1,3. Let me see how things go from here.
Up Topic The Rybka Lounge / Computer Chess / Jim Ablett's build?? or Author's build??

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill