Crafty-23.1 bench --- lower?

[zullil]

No matter how I optimize, my Crafty-23.1 builds give significantly lower bench marks than I get from 23.0, by 25% or more. What am I missing this time? Anyone else seeing this?

Code: Select all

Crafty v23.1 (8 cpus)

White(1): bench
Running benchmark. . .
......
Total nodes: 164955544
Raw nodes per second: 11145644
Total elapsed time: 14.80

[zullil]

Still puzzled.

[jarkkop]

Why?
They have changed how search functions and added some code that takes more time, but is more accurate. There a lot of changes that account for about 100 ELO.

[jarkkop]

Here is what I get with Q8400 @ 3200GHz (four threads)
If you have 8 real cores, then something is wrong

Crafty v23.1 (1 cpus)

White(1): hash 128M
hash table memory = 128M bytes.
White(1): hashp 64M
pawn hash table memory = 64M bytes.
White(1): mt 4
max threads set to 4.
White(1): bench
Running benchmark. . .
......
Total nodes: 108877958
Raw nodes per second: 9997975
Total elapsed time: 10.89
White(1):

[zullil]

Here is what I get with Q8400 @ 3200GHz (four threads)
If you have 8 real cores, then something is wrong

I do have 8 cores, so something is wrong. Now to find out what!

Thanks for the data.

[zullil]

Here is what I get with Q8400 @ 3200GHz (four threads)
If you have 8 real cores, then something is wrong

Crafty v23.1 (1 cpus)

White(1): hash 128M
hash table memory = 128M bytes.
White(1): hashp 64M
pawn hash table memory = 64M bytes.
White(1): mt 4
max threads set to 4.
White(1): bench
Running benchmark. . .
......
Total nodes: 108877958
Raw nodes per second: 9997975
Total elapsed time: 10.89
White(1):

I have two 2.26GHz Quad-Core Intel Xeon "Nehalem" processors in a Mac Pro. After updating to the latest Intel C compiler for OS X and using pgo, I now get

Code: Select all

White(1): bench
Running benchmark. . .
......
Total nodes: 115079525
Raw nodes per second: 17841786
Total elapsed time: 6.45
White(1): bench
Running benchmark. . .

which seems more like it!

[Martin Thoresen]

Sorry to bump this old thread, but wanted to add my own bench results.

Code: Select all

Crafty v23.1 JA <4 cpus>

White<1>: bench
Running benchmark. . .
......
Total nodes: 1171155125
Raw nodes per second: 19018689
Total elapsed time: 6.16
White<1>:

[Peter Skinner]

found computer opening book file [./bookc.bin].

Initializing multiple threads.
System is SMP, not NUMA.
max threads set to 4.
EGTB access enabled
using tbpath=./TB
5 piece tablebase files found
14043kb of RAM used for TB indices and decompression tables
hash table memory = 2048M bytes.
pawn hash table memory = 128M bytes.
EGTB cache memory = 32M bytes.
resign after 5 consecutive moves with score < -6.


Crafty v23.1 (4 cpus)

White(1): bench
Running benchmark. . .
......
Total nodes: 113016951
Raw nodes per second: 12244523
Total elapsed time: 9.23
White(1):

This is on my Phenom x4 920 system. (I never over clock, same compile from my website)