TalkChess.com

I measure the percentage of transposition table lookups (probes) that get a hit - a hit meaning there is info about the requested position already stored in the table. And I'm getting around 17-20% hits on average.

When I run other engines in Arena they show 99%-100% in the Hash column. So I'm wondering if that percentage is hit rate as described above, and if so how does one get hit rate so high? I mean, positions don't repeat that often within the search tree do they?

The value displayed in Arena is hash full, not hit rate. I usually get 25-30% hash hits midgame if I give sufficient space for the table.

Ah, hash full makes much more sense with those percentages. Thanks for the clarification.

Anyone know what kind of hit rate most top engines get these days? With say 1GB table size?

zd3nik wrote:Ah, hash full makes much more sense with those percentages. Thanks for the clarification.

Anyone know what kind of hit rate most top engines get these days? With say 1GB table size?

Depends on the position. Normal middle game maybe 20-30% max. In endgames like fine #70, >75%...

I did a test on Andscacs with current developing version.
Go depth 25 from start position

Thanks for the feedback guys. It's good to have a sanity check every once in a while.

cdani wrote:I did a test on Andscacs with current developing version.
Go depth 25 from start position

Code: Select all

hash            nodes       stored   not found       found  % found over found+not found
128 MB	  53.425.849   65.716.083  36.862.078  18.983.223  34%
256 MB	  64.437.112   78.582.936  43.534.378  23.853.827  35%
1024 MB	118.011.879  143.221.524  77.990.897  45.678.716  37%
4096 MB	 74.634.324   88.762.678  48.235.382  29.907.519  38%

what is "found"? Signature match? Signature match + draft sufficient to use? Signature match + draft sufficient + bound good enough to stop search???

bob wrote:

cdani wrote:I did a test on Andscacs with current developing version.
Go depth 25 from start position

Code: Select all

hash            nodes       stored   not found       found  % found over found+not found
128 MB	  53.425.849   65.716.083  36.862.078  18.983.223  34%
256 MB	  64.437.112   78.582.936  43.534.378  23.853.827  35%
1024 MB	118.011.879  143.221.524  77.990.897  45.678.716  37%
4096 MB	 74.634.324   88.762.678  48.235.382  29.907.519  38%

what is "found"? Signature match? Signature match + draft sufficient to use? Signature match + draft sufficient + bound good enough to stop search???

"Found" is just signature match.

cdani wrote:

bob wrote:

cdani wrote:I did a test on Andscacs with current developing version.
Go depth 25 from start position

Code: Select all

hash            nodes       stored   not found       found  % found over found+not found
128 MB	  53.425.849   65.716.083  36.862.078  18.983.223  34%
256 MB	  64.437.112   78.582.936  43.534.378  23.853.827  35%
1024 MB	118.011.879  143.221.524  77.990.897  45.678.716  37%
4096 MB	 74.634.324   88.762.678  48.235.382  29.907.519  38%

what is "found"? Signature match? Signature match + draft sufficient to use? Signature match + draft sufficient + bound good enough to stop search???

"Found" is just signature match.

And what is the reason for the small difference between "nodes" and "found + not found"?

Sven Schüle wrote: And what is the reason for the small difference between "nodes" and "found + not found"?

Sorry, I didn't explain. Nodes are the reported nodes at the end of the search, so they are not directly related to Hash table.