Preliminary 16-core Test Results (no hypothesis)
W7 Ultimate, 64-bit
Fritz Chess Benchmark: 46.57
50 move, Silver Test Suite/book, colors reversed.
E5-2687W (16 physical cores); no HT.
SF idle threads sleep; Min. split depth default;
H contempt 0.
No PB or TB.
Hash 2048
SF010214-16 v. H4B-16, Blitz 1m+1s 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
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
SF idle threads sleep.
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
SF010214-16 v. H4B-1, Blitz 1m+1s 0
1 Stockfish 020114 64 SSE4.2 +174 +41/=35/-4 73.13% 58.5/80
2 Houdini 4 Pro x64B -174 +4/=35/-41 26.88% 21.5/80
This match was stopped at 80 due to technical reasons, so the inverse test will also be 80 games.
SF010214-1 v. H4B-16, Blitz 1m+1s 0
1 Houdini 4 Pro x64B +205 +47/=30/-4 76.54% 62.0/81
2 Stockfish 020114 64 SSE4.2 -205 +4/=30/-47 23.46% 19.0/81
These are a prelude to Vincent's request for a 16 v 8, 5+3 test.
ouachita wrote:I'll run a few later on with SF 32 threads just to see if things change noticeably.
Not to hijack your thread, but did you find any info about a good value for Min Split Depth when Threads=16? If you are using the default, then I believe you're using Min Split Depth=7 (which might be sub-optimal for 16 threads).
No. I asked Marco who didn't reply, so he might not know either or might not have picked up his messages. Applying Clemens' theory, it should be 9 here.
My system is the two socket 12 cores with HT off, old SR-2.
I just ran the Autotune for Houdini 4 and my best min split depth is 12 same as with Houdini 3.
I wonder what makes one program want a min split depth 9 and another program wants min split depth 19?
Also Bobby, I wonder if your Fritz bench can use all of your cores, some of the Fritz bench tests were limited in core usage.
kgburcham wrote:I wonder if your Fritz bench can use all of your cores, some of the Fritz bench tests were limited in core usage.kgburcham
Kim,
You may not remember, but I consulted with you via the Rybka forum before and after purchasing this 2687W workstation. As to your question above, when I run the Fritz Chess Benchmark test in DF 13, it appears to use all 16 cores but how can I actually determine whether or not and how well it is using all 16?
Regarding split depth, I suppose I would have to run matches or tournaments to determine the best for SF if such tests have not already been performed, and I doubt they have using 16 cores.
ouachita wrote:Regarding split depth, I suppose I would have to run matches or tournaments to determine the best for SF if such tests have not already been performed, and I doubt they have using 16 cores.
Nodes/sec comparison for 1 or 2 minutes is probably enough (average on 5 runs for each config) !
ouachita wrote:I'll run a few later on with SF 32 threads just to see if things change noticeably.
Not to hijack your thread, but did you find any info about a good value for Min Split Depth when Threads=16? If you are using the default, then I believe you're using Min Split Depth=7 (which might be sub-optimal for 16 threads).
No. I asked Marco who didn't reply, so he might not know either or might not have picked up his messages. Applying Clemens' theory, it should be 9 here.
hmmm, I cannot remember this theory !?
I never thought abot SD 9 for Stockfish.
In my tournaments I leave it to zero, so I think the engine will choose SD7 as default value.
I had choosen SD 12 for Houdini 4 in former tournaments. But I steped back from that and leave all default now in my tours.
ouachita wrote:Regarding split depth, I suppose I would have to run matches or tournaments to determine the best for SF if such tests have not already been performed, and I doubt they have using 16 cores.
Nodes/sec comparison for 1 or 2 minutes is probably enough (average on 5 runs for each config) !
ouachita wrote:Regarding split depth, I suppose I would have to run matches or tournaments to determine the best for SF if such tests have not already been performed, and I doubt they have using 16 cores.
Nodes/sec comparison for 1 or 2 minutes is probably enough (average on 5 runs for each config) !
Avg. - 22,000
thumbs up or down?
It's specific to your system, so you've to test 8 and 9 too (may be 10 ?) to find the fastest "min split depth" value.
Hugo wrote:
hmmm, I cannot remember this theory !?
I never thought about SD 9 for Stockfish.
Clemens
3 Champs Reloaded
Engines:
Houdini 4.0 pro, split depth 12
Stockfish 241113, sleeping threads idle ON, min split depth 7
Am I missing something?
Min Split Depth = 7 is the default that SF will use whenever Threads ≥ 8. That number was adopted based on testing I did long ago on my then new 8-core Mac Pro.
SF has changed a lot since then. I'm about to do a bit of testing of the latest SF (still on my now old 8-core machine).