Compared to Rybka, Stockfish should score better in 32-bit!

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

ernest
Posts: 2041
Joined: Wed Mar 08, 2006 8:30 pm

Compared to Rybka, Stockfish should score better in 32-bit!

Post by ernest »

If you compare the 64-bit speed to the 32-bit speed
(infinite analysis of the start_position, 1-cpu/thread for reproducibility),
you get:
+70% for Rybka 4
+25% for Stockfish 1.9

So compared to Rybka, Stockfish should perform quite a bit better in 32-bit!
In the rating lists, the Rybka-Stockfish gap should be around 30 Elo less in 32-bit than in 64-bit.
Jouni
Posts: 3283
Joined: Wed Mar 08, 2006 8:15 pm

Re: Compared to Rybka, Stockfish should score better in 32-b

Post by Jouni »

Seems to be true, because in CSS forum:

Stockfish 1.9.1 JA v/s Rybka 3 58 - 42 (+ 39,= 38,- 23) !

But I don't understand why test is done with 1 cpu (no ponder) in dual PC??

Jouni
swami
Posts: 6640
Joined: Thu Mar 09, 2006 4:21 am

Re: Compared to Rybka, Stockfish should score better in 32-b

Post by swami »

Very True.

Computer here is 32 bits and this is precisely why I see such a good improvement from Stockfish 1.9 in STS test. :)

Best wishes,
Swami
Frank Quisinsky
Posts: 6808
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Compared to Rybka, Stockfish should score better in 32-b

Post by Frank Quisinsky »

Hi Jouni,

look here, SWCR results:

Code: Select all

   5 Stockfish 1.8.0 JA x64       2901   18   18  1240   74%  2718   33% 
   9 Stockfish 1.7.1 JA x64       2894   19   18  1120   76%  2704   34% 
  10 Stockfish 1.7.1 JA w32       2874   18   18  1200   75%  2690   31% 
  11 Stockfish 1.8.0 JA w32       2863   20   20   920   74%  2697   37%
Between SF 1.71 x64 to w32 = +20 ELO
Between SF 1.80 x64 to w32 = +38 ELO

Now SF 1.9.0 / 1.9.1 are available.

SF 1.9.0 JA w32 = 336.384 bytes
SF 1.9.1 JA w32 = 206.336 bytes

I think the different results for the latest w32 versions have to do with the compiles from Jim.

Same for SF 1.6.0 up to 1.6.3
The Stockfish team sayed ... same playing level, no changes ...

But look here, again my SWCR results:

Code: Select all

 15 Stockfish 1.6.3 JA w32       2827   18   18  1080   71%  2680   36% 
 20 Stockfish 1.6.0 JA w32       2802   19   18   960   68%  2687   39%
A bit confusion for myself and mental confusion in SF w32 results for the SWCR visitors.

Not sure, but I believe that the compiles Jim made are variable for the 32bit versions.

Version 1.9.1 not tested yet in SWCR. Many work with the IPP family test. SF 1.9.1 w32 / x64 comes a little bit later.

Best
Frank
Jouni
Posts: 3283
Joined: Wed Mar 08, 2006 8:15 pm

Re: Compared to Rybka, Stockfish should score better in 32-b

Post by Jouni »

With 64b and dual I got in full STS 1300 position / 10s:

Rybka 3 1164 / 42m
Stockfish 1.9 1099 / 54m

They should be very equal.

Jouni