Crafty 19.18

[Ciro Vignotto]

Hi all,
Could anyone try to compile latest Crafty version, please?
TIA, Ciro

[Jim Ablett]

Hi Ciro,

Here it is,
Compiled exe from latest 1918 source code.
GCC 3.3 - mmx optimized.

http://homepages.tesco.net/henry.ablett ... 918mmx.zip

Regards,
Jim.

[Jim Ablett]

If anyone would like a P4 or Athlon optimized build,
contact me.

Jim.

[Pablo]

Hi Jim:
Why this Crafty19.18 version compiled by you, weight about 200Kb instead of all previous versions (19.17,19.15 etc.) compiled by different people that weight around 800 to 900 KB
Cheers,
Pablo Urzua

[Jim Ablett]

Hello Pablo,

There are a few reasons for this.

1. Using GCC with certain selected optimizations reduces to compiled code size considerably. e.g high optimization (-O3) is actually worst than medium (-O2, it produces a lot of excessive in-line stuff.

2. Other compilers can add a lot of bloat to the executable file, Microsoft, Intel etc.

3. I compress the file after compilation with an executable packer which halves it's size, (in this case from 400kb to 200kb)


4. I'm just a cool compiler I guess!


Jim.

[F.Huber]

Hello Pablo,

There are a few reasons for this.

1. Using GCC with certain selected optimizations reduces to compiled code size considerably. e.g high optimization (-O3) is actually worst than medium (-O2, it produces a lot of excessive in-line stuff.

2. Other compilers can add a lot of bloat to the executable file, Microsoft, Intel etc.

3. I compress the file after compilation with an executable packer which halves it's size, (in this case from 400kb to 200kb)


4. I'm just a cool compiler I guess!


Jim.

Sorry Jim, but I have to add another reason:

5. Your executable is dramatically slower than other ones (e.g. from Dann Corbit), at least on my Celeron/400 -
it gives only 75kN/s compared to about 115kN/s of other compilers!

Regards,
Franz.

[Jim Ablett]

Yes, that's another reason!

I think a P2 optimized version will run a bit faster on your machine.

I just wanted to compile a general (non-processor specific) mmx version for universal compatibility reasons which will always be slower.

Jim.

[Dann Corbit]

Here is a MS VC++ version:
ftp://cap.connx.com/pub/chess-engines/n ... y-1918.exe

Here is the same thing compressed with bzip2:
ftp://cap.connx.com/pub/chess-engines/n ... 18.exe.bz2

[Jim Ablett]

Try this one (optimized for P2/Athlon-XP)

http://homepages.tesco.net/henry.ablett/crafty1918.zip

Jim.

[Dann Corbit]

CPU Identification utility v1.10 (c) 1997-2004 Jan Steunebrink
------------------------------------------------------------------------------
CPU Vendor and Model: AMD Athlon 64 2800+-3700+
Internal CPU speed : 2199.4 MHz (using internal Time Stamp Counter)
Clock Multiplier : Available only in Real Mode!
CPU-ID Vendor string: AuthenticAMD
CPU-ID Name string : AMD Athlon(tm) 64 Processor 3400+
CPU-ID Signature : 0F4A
???+- Stepping or sub-model no.
??+- Model: Indicates CPU Model and 486 L1 cache mode
?+- Family: 4=486, Am5x86, Cx5x86
? 5=Pentium, Nx586, Cx6x86, K5/K6, C6, mP6
? 6=PentiumPro/II/III, CxMII/III, Athlon, C3
? F=Pentium4, Athlon64
+- Type: 0=Standard, 1=Overdrive, 2=2nd Dual Pentium
Current CPU mode : Protected
Internal (L1) cache : Enabled in Write-Back mode

1 GB Ram

DC build:
EPD Kit revision date: 1996.04.21
found computer opening book file [./bookc.bin].
ERROR "krafty fischer " is unknown rc-file option
hash table memory = 96M bytes.
pawn hash table memory = 12M bytes.
EGTB cache memory = 8M bytes.
draw score set to 0.00 pawns.
choose from book moves randomly (using weights.)
choose from 5 best moves.
book learning enabled
result learning enabled
position learning enabled
resign after 5 consecutive moves with score < -9.
5 piece tablebase files found
13980kb of RAM used for TB indices and decompression tables

Crafty v19.18

White(1): bench
Running benchmark. . .
......
Total nodes: 89179462
Raw nodes per second: 1393429
Total elapsed time: 64
SMP time-to-ply measurement: 10.000000
White(1):

JA build:
found computer opening book file [./bookc.bin].
ERROR "krafty fischer " is unknown rc-file option
hash table memory = 96M bytes.
pawn hash table memory = 12M bytes.
EGTB cache memory = 8M bytes.
draw score set to 0.00 pawns.
choose from book moves randomly (using weights.)
choose from 5 best moves.
book learning enabled
result learning enabled
position learning enabled
resign after 5 consecutive moves with score < -9.
5 piece tablebase files found
13980kb of RAM used for TB indices and decompression tables

Crafty v19.18

White(1): bench
Running benchmark. . .
......
Total nodes: 90745371
Raw nodes per second: 1226288
Total elapsed time: 74
SMP time-to-ply measurement: 8.648649
White(1):

[Bryan Hofmann]

I just finished reading the Crafty newsgroup and I would recommend holding off on running this new version of Crafty. It appears that there is a rather large memory leak with the New EGTB code. Eugene is aware of this and will be working on it to fix this issue.


Bryan

[Jim Ablett]

'Dann's the man' when it comes to fast crafty compiles.

Faster isn't always better though, mine thinks slower, so it makes less tactical blunders! ...ha ha


I knew this remark would come back to haunt me!

I'm just a cool compiler I guess!

Jim

[Dann Corbit]

It sounds like it is broken now anyway, with some sort of memory leak in the EGTB code.

But if you do not have tablebase files, it should be OK.

[Norm Pollock]

Crafty 19.18 works fine under fritz8 with 3-4-5 egtbs. The only problem I see is that it doesn't play chess quite as well as Crafty 19.13.

My testing so far is through 52 games at 4'+2", on amd 3000 xp pro, 128M each, 32M egtb, same book (no learning) based on my pgn file.

After 52 games: Crafty 19.13: 18 wins, 12 losses, 22 draws, 55.77%.

[Norm Pollock]

follow-up to the above post:

I finished the 60 game match between crafty19.13 v crafty19.18. The gap closed a bit: crafty 19.13: 20 wins, 15 losses, 25 draws, 54.17%.

I used the DC version for 19.13, and the PS sse version for 19.18. On my machine, for 19.18, the PS sse version has a 10% higher nps rate (1,209,520 v 1,079438nps) than the DC version. So I assume that if I had the PS sse version for 19.13, the results would have been even better for 19.13.

[Bryan Hofmann]

follow-up to the above post:

I finished the 60 game match between crafty19.13 v crafty19.18. The gap closed a bit: crafty 19.13: 20 wins, 15 losses, 25 draws, 54.17%.

I used the DC version for 19.13, and the PS sse version for 19.18. On my machine, for 19.18, the PS sse version has a 10% higher nps rate (1,209,520 v 1,079438nps) than the DC version. So I assume that if I had the PS sse version for 19.13, the results would have been even better for 19.13.

One of the reasons for the differences in NPS is due to the fact of the crafty compile time flag options. Dan uses the detect draw which does slow things down abit and does not use the futility pruning where Peter's complie does.

[Bryan Hofmann]

I have updated my new site with the 19.18 compiles. If you are playing with 5 man EGTBs and below you should not experience the memory leak issue as it appears to only effect those using the newer 6 man tables.

http://www.freewebs.com/bkhofmann/

Below are the benchmarks on a AMD XP 3000+ running with default settings (ie no crafty.rc file)

Crafty v19.18 BH

White(1): ben
Running benchmark. . .
......
Total nodes: 96761642
Raw nodes per second: 1290155
Total elapsed time: 75
SMP time-to-ply measurement: 8.533333

Crafty v19.18 PS

White(1): ben
Running benchmark. . .
......
Total nodes: 96761642
Raw nodes per second: 1273179
Total elapsed time: 76
SMP time-to-ply measurement: 8.421053

Crafty v19.18 DC

White(1): ben
Running benchmark. . .
......
Total nodes: 97149438
Raw nodes per second: 1142934
Total elapsed time: 85
SMP time-to-ply measurement: 7.529412

Crafty v19.18 JA

White(1): ben
Running benchmark. . .
......
Total nodes: 96761642
Raw nodes per second: 967616
Total elapsed time: 100
SMP time-to-ply measurement: 6.400000