CCRL 40/4: Crafty 21.5 PS 64-bit

[Kirill Kryukov]

Recently I did some testing for Crafty 21.5 PS 64-bit. "PS" means Peter Skinner's compile downloaded from here. It played 976 games so far and obtained rating of 2649 (+19 −19), which makes it #22 free single-CPU engine (under CCRL 40/4 conditions).

Free single-CPU rating list and cross-table (my version)
All results of Crafty 21.5 PS 64-bit

Crafty used 1 CPU in this test. It should be much stronger with 2 or more CPUs, but I am interested in single-CPU performance at the moment.

Also several engines around it are very near to each other so the ranking is not very reliable and may easily change in future with more games:

Code: Select all

 Rank                 Engine                  ELO   +    -   Score  AvOp  Games
   20 The Baron 1.8.1                        2654  +21  -21  45.5%  +32.3   789
   21 Smarthink 0.17a                        2651  +22  -22  45.9%  +28.1   702
   22 Crafty 21.5 PS 64-bit                  2649  +19  -19  40.6%  +69.3   976
   23 Colossus 2006f                         2646  +17  -17  47.0%  +20.7  1181
   24 Jonny 2.83 32-bit                      2645  +17  -17  41.5%  +64.1  1216

Comparing to other Crafty versions it looks like 21.5 PS 64-bit is at least stronger than other single-CPU versions we tested, but some other verisons have small number of games to make good comparison.

Comparison of all Crafty versions that played in CCRL 40/4

Note that Crafty 21.5 source is not available yet, that's why it is marked with green color in our list (open source engines are in orange color), and it does not participate in the open source rating list.

Good work from the Crafty team, although I was hoping for at least top 20. (Note that Strelka is even not counted here yet). Looking forward to see stronger future versions, and also to see the source of course.

Best,
Kirill

[Dirt]

Comparison of all Crafty versions that played in CCRL 40/4

So it looks like 20.14 is stronger than 21.5. Good to know.

[Spock]

Comparison of all Crafty versions that played in CCRL 40/4

So it looks like 20.14 is stronger than 21.5. Good to know.

No it's not - you're looking at the 2CPU version of 20.14

[Uri Blass]

Comparison of all Crafty versions that played in CCRL 40/4

So it looks like 20.14 is stronger than 21.5. Good to know.

No it's not - you're looking at the 2CPU version of 20.14

No
We have:
1 Crafty 20.14 32-bit 2CPU 2713 +32 −32 44.4% +38.4 34.2% 319
100.0%
Crafty 21.5 PS 64-bit 2649 +19 −19 40.6% +69.3 30.7% 976
90.5%
Crafty 20.14 32-bit 2628 +25 −25 38.8% +86.4 22.8% 628
92.3%



you can see that 21.5 used 64 bits when 20.14 used only 32 bits.
64 bits for bitboard engines should give a significant speed improvement so I guess based on the data that 20.14 is better than 21.5

Uri

[Graham Banks]

Comparison of all Crafty versions that played in CCRL 40/4

So it looks like 20.14 is stronger than 21.5. Good to know.

No it's not - you're looking at the 2CPU version of 20.14

No
We have:
1 Crafty 20.14 32-bit 2CPU 2713 +32 −32 44.4% +38.4 34.2% 319
100.0%
Crafty 21.5 PS 64-bit 2649 +19 −19 40.6% +69.3 30.7% 976
90.5%
Crafty 20.14 32-bit 2628 +25 −25 38.8% +86.4 22.8% 628
92.3%



you can see that 21.5 used 64 bits when 20.14 used only 32 bits.
64 bits for bitboard engines should give a significant speed improvement so I guess based on the data that 20.14 is better than 21.5

Uri

I can assure you that Crafty 21.5 32-bit 1CPU is certainly stronger than Crafty 20.14 32-bit 1CPU at 40/40 and by a reasonably nice margin from memory.

[Dirt]

...you can see that 21.5 used 64 bits when 20.14 used only 32 bits.
64 bits for bitboard engines should give a significant speed improvement so I guess based on the data that 20.14 is better than 21.5

Uri

Yes, that was the basis for my comment.