(Multi-threaded) Cfish-dev is 37% faster than Stockfish-dev??

[zullil]

Linux workstation with dual Intel(R) Xeon(R) CPU E5-2687W v3 @ 3.10GHz (giving 20 physical cores).

Was really surprised by the following results. Cfish binds threads to cores as part of its NUMA handling, but I don't recall Cfish being so much faster than Stockfish in the past.

With 20 threads (typical results, but not deterministic):

$ ./cfish bench 4096 20 18 >> \dev\null

===========================
Total time (ms) : 5383
Nodes searched : 182588585
Nodes/second : 33919484

$ ./stockfish bench 4096 20 18 >>/dev/null

===========================
Total time (ms) : 7891
Nodes searched : 195008953
Nodes/second : 24712831

Cfish-dev is about 37% faster with twenty threads.

With 1 thread (deterministic):

$ ./cfish bench >>/dev/null

===========================
Total time (ms) : 1905
Nodes searched : 3897263
Nodes/second : 2045807

$ ./stockfish bench >>/dev/null

===========================
Total time (ms) : 2092
Nodes searched : 3897263
Nodes/second : 1862936

Cfish-dev is about 10% faster with one thread.