16 Core (E5-2687W) Matches and Tournaments

[ouachita]

Code: Select all

SF020114-8C v. H4B-16C, Blitz 2m  0

49/100 games

Houdini 4 Pro x64B-Stockfish 020114 64 SSE4.2: 29.0-20.0: +18/=22/-9: 59.18%

Match will be finished shortly.

[ouachita]

Code: Select all

 
100 Game Matches

H4B-16C v. SF020114-8C, Blitz 2m  0

Houdini 4 Pro x64B-16 - Stockfish 020114 64 SSE4.2-8   57.0 - 43.0   +34/=46/-20    57.00%

SF010214-16 v. H4B-16, Blitz 3m  0
                                      
1   Stockfish 020114 64 SSE4.2    +3  +20/=61/-19 50.50%   50.5/100
2   Houdini 4 Pro x64B            -3  +19/=61/-20 49.50%   49.5/100

SF010214-16 v. H4B-16, Blitz 5m  0
                                      
1   Stockfish 020114 64 SSE4.2   +49  +29/=56/-15 57.00%   57.0/100
2   Houdini 4 Pro x64B           -49  +15/=56/-29 43.00%   43.0/100

The bases for these tests are the same except for time and the number of cores for each engine.

Hypothesis: ELO increases by an as-yet unquantified amount as the number of physical cores increases; or, ELO decreases by an as-yet unquantified amount as the number of physical cores decreases.

It also seems that both engines' time management is better with fixed game times, i.e., 2 min., 3 min., 5 min., etc., rather than min. + X.

[ouachita]

Code: Select all

100 Game Matches - 16 cores

Surprisingly, H4B is stronger than SF 020114 at both 16 v 8 and 8 v 16 with this test basis:

W7 Ultimate, 64-bit 
Fritz Chess Benchmark: 47.97 (23,032 kN/s) 
50 move, Silver Test Suite/book, colors reversed. 
E5-2687W (16 physical cores); HT disabled. 
No PB or TB. 
Hash 2048
2 min. + 0

H4B-16 v. SF020114-8, 2m  0   

1   Houdini 4 Pro x64B 57.0 (+34/=46/-20    57.00%)
2   Stockfish 020114 64 SSE4.2 43.0

SF020114-16C v. H4B-8C, 2m  0
                                      
1   Houdini 4 Pro x64B            +7  +22/=58/-20 51.00%   51.0/100
2   Stockfish 020114 64 SSE4.2    -7  +20/=58/-22 49.00%   49.0/100

3 + 0 match is running now.

[ouachita]

Code: Select all

Preliminary 16-core Test Results (no hypothesis)

#1 SF010214-16C v. H4B-16C, 1+1  0
                                      
1   Houdini 4 Pro x64B           +28  +29/=50/-21 54.00%   54.0/100
2   Stockfish 020114 64 SSE4.2   -28  +21/=50/-29 46.00%   46.0/100

#2 SF010214-32T v. H4B-16C, 1+1  0
                                      
1   Houdini 4 Pro x64B            +7  +21/=60/-19 51.00%   51.0/100
2   Stockfish 020114 64 SSE4.2    -7  +19/=60/-21 49.00%   49.0/100

#3 SF010214-16C v. H4B-16C, 3m  0
                                      
1   Stockfish 020114 64 SSE4.2    +3  +20/=61/-19 50.50%   50.5/100
2   Houdini 4 Pro x64B            -3  +19/=61/-20 49.50%   49.5/100
SF idle threads sleep.

#4 SF010214-16C v. H4B-16C, 5m  0
                                      
1   Stockfish 020114 64 SSE4.2   +49  +29/=56/-15 57.00%   57.0/100
2   Houdini 4 Pro x64B           -49  +15/=56/-29 43.00%   43.0/100

#5 H4B-16C v. SF020114-8C, 2m  0    

1   Houdini 4 Pro x64B 57.0 (+34/=46/-20    57.00%) 
2   Stockfish 020114 64 SSE4.2 43.0

#6 SF020114-16C v. H4B-8C, 2m  0
                                      
1   Houdini 4 Pro x64B            +7  +22/=58/-20 51.00%   51.0/100
2   Stockfish 020114 64 SSE4.2    -7  +20/=58/-22 49.00%   49.0/100

#7 SF020114-16C v. H4B-8C, 3m  0
                                      
1   Stockfish 020114 64 SSE4.2   +10  +26/=51/-23 51.50%   51.5/100
2   Houdini 4 Pro x64B           -10  +23/=51/-26 48.50%   48.5/100

I feel sure H would dominate a H16 v SF 8 match at 3min, as you can see SF was gained only a few ELO when H drops 8 cores. 

All of this is headed toward a 5min round robin between SF, H, and K.

Generally, I'm sure you knew and can see that one should not be optimistic while bringing 1, 2, 4 or 8 core machines to a STC match versus 16 cores. I've tested 16 v 1, 16 v 2, etc. and the one sided results are not surprising.

[ouachita]

Code: Select all

Addendum:

16 cores v 16 cores

SF050114IP-16C v. H4B-16C, 5m  0
                                        
1   Houdini 4 Pro x64B             +24  +22/=63/-15 53.50%   53.5/100
2   Stockfish 050114IP 64 SSE4.2   -24  +15/=63/-22 46.50%   46.5/100

W7 Ultimate, 64-bit 
50 move, Silver Test Suite/book, colors reversed. 
E5-2687W (16 physical cores); HT disabled. 
No PB or TB. 
Hash 2048

H4B continues to be better than SF from 1+1 thru 5+0 with these test bases.

[ouachita]

Code: Select all

16 cores v 16 cores
SF080114-16C v. KTCECr-16C, Blitz 2m  0

W7 Ultimate, 64-bit 
50 move, Silver Test Suite/book, colors reversed. 
E5-2687W (16 physical cores); HT disabled. 
No PB or TB. 
Hash 2048
                                      
1   Komodo TCECr 64-bit          +28  +33/=42/-25 54.00%   54.0/100
2   Stockfish 080114 64 SSE4.2   -28  +25/=42/-33 46.00%   46.0/100

I ran a 5m test last night and the result was in K's favor. I was so surprised that I ran this test. On 16 cores at these TC, K is simply better. I'm not certain why I was surprised.

[ernest]

I'm not certain why I was surprised.

Well, when you realize that for this 100-game match,
your 95% error-bar is ±7.6% or ±53 Elo, you can't be surprised!

[ouachita]

Code: Select all

1/18/14
1m, 5 engine, Round Robin
Openings DB: 5 Pivotal Positions from Bobby Johnson's ICCF games (wins), Alternate colors
8 cores/threads
PB On
4096 hash

                                  1          2          3          4          5          
1   Stockfish 050114IP 64 SSE4.2  ********** ½½½½½1½½01 ½½½½10½½½½ ½½½½½10½½1 0½1½½11111  23.5/40
2   Houdini 4 Pro x64B            ½½½½½0½½10 ********** ½0½1½10½½1 ½½½10½½0½1 ½½½½11½½½1  21.5/40
3   Stockfish 090114IP 64 SSE4.2  ½½½½01½½½½ ½1½0½01½½0 ********** ½½½01½1½10 ½1½½½½½10½  20.5/40
4   Stockfish 080114 64 SSE4.2    ½½½½½01½½0 ½½½01½½1½0 ½½½10½0½01 ********** 00½½10½½½½  18.0/40
5   Komodo TCECr 64-bit           1½0½½00000 ½½½½00½½½0 ½0½½½½½01½ 11½½01½½½½ **********  16.5/40

Study this during today's NFL championship games.

[ouachita]

or this one:

Code: Select all

19Jan14

1m, 5 engine, Round Robin 
Openings DB: 5 Pivotal Positions from Bobby Johnson's ICCF games (wins), Alternate colors 
6 cores/threads 
PB On 
6144 hash 
                                  1          2          3          4          5          
1   Stockfish 080114 64 SSE4.2    ********** 0001½10001 ½½0½01011½ ½1½½1½1½½½ ½11½½½0½1½  21.0/40
2   Komodo TCECr 64-bit           1110½01110 ********** ½10½½½10½1 ½0½½100½1  ½½0½½1½½0   20.0/38
3   Stockfish 050114IP 64 SSE4.2  ½½1½10100½ ½01½½½01½0 ********** 0½1001½1   10½11½½1½   19.5/37  363.00
4   Stockfish 090114IP 64 SSE4.2  ½0½½0½0½½½ ½1½½011½0  1½0110½0   ********** 11½1½½01½1  19.5/37  356.50
5   Houdini 4 Pro x64 B           ½00½½½1½0½ ½½1½½0½½1  01½00½½0½  00½0½½10½0 **********  15.0/38

[ouachita]

Code: Select all

1/18/14 

3m, 5 engine, Round Robin 
Openings DB: 5 Pivotal Positions from Bobby Johnson's ICCF games (wins), Alternate colors 
8 cores/threads 
PB On 
Various non-default engine parameters
4096 hash 


                                  1          2          3          4          5          
1   Komodo TCECr 64-bit           ********** ½½½½½½1½1½ 10½½½½½0½½ 0½½0½1½11½ 11½1½1½½11  24.0/40
2   Stockfish 090114IP 64 SSE4.2  ½½½½½½0½0½ ********** ½1½½½½½½½1 ½½1½½101½½ ½½½01½½11½  22.0/40
3   Stockfish 050114IP 64 SSE4.2  01½½½½½1½½ ½0½½½½½½½0 ********** 1½½10½10½½ ½½010½½½0½  19.0/40
4   Stockfish 080114 64 SSE4.2    1½½1½0½00½ ½½0½½010½½ 0½½01½01½½ ********** 001½½0011½  17.5/40  360.25
5   Houdini 4 Pro x64B            00½0½0½½00 ½½½10½½00½ ½½101½½½1½ 110½½1100½ **********  17.5/40  346.25