Perft(13), final fifth

[Daniel Shawul]

The earlier calculation of perft(11) for 1 e4 e5 using 56 bits of hash was 22,653,878,987,990,237 and this is apparently INCORRECT.

The main run, using a 120 bit hash, gives the most likely by far correct number of 22,653,878,988,011,879.

Close, but close isn't good enough in this game.

Wow that was scary! With such a risk of corrupting a year's work, one wonders how muchthe benefit of hashing is? I understand you split the tree and do 8-ply perfts so you remove transpositions there. Maybe not using hashing in the rest of the plies doesn't hurt much...

[sje]

The earlier calculation of perft(11) for 1 e4 e5 using 56 bits of hash was 22,653,878,987,990,237 and this is apparently INCORRECT.

The main run, using a 120 bit hash, gives the most likely by far correct number of 22,653,878,988,011,879.

Close, but close isn't good enough in this game.

Wow that was scary! With such a risk of corrupting a year's work, one wonders how muchthe benefit of hashing is? I understand you split the tree and do 8-ply perfts so you remove transpositions there. Maybe not using hashing in the rest of the plies doesn't hurt much...

Actually, the code splits the ply zero move set at the root. For each ply zero move, there is a separate thread and all the threads share the same WTM transposition table and the same BTM transposition table.

With an extra 64 bits of hash in addition to the base 56 bits, I am very confident that there wont be any false positives in a hundred years of running at currant rates.

[sje]

The run has been producing draft 11 results at the rate of about two per day for a while, and this is about twice as fast as the overall rate to date. There are now only 32 draft 11 calculations remaining, so the End Is Near.

There are 16 "hard" draft 11 calculations; these are ply two position where both sides have advanced a center pawn. All 16 of these have completed and all but one of these rank in the top 16 by count draft 11 results.

There are only two draft 11 calculations left for the 1 h4 draft 12 result, so that should show up in a day or so. Once the 1 h4 number appears, there will be only three threads running on the four core CPU.

The remaining draft 11 positions are:

1 d4 f6/f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e3 f6/f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e4 f6/f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 h4 Nf6/Nh6

[Sorry for all the edits. I have to take a lot of meds these days.]

[sje]

Perft(12) for 1 h4 is 80,876,962,213,785,922 and this is the 17th draft 12 calculation made by the run so far; three more to go. There are now three threads (1 d4, 1 e3, 1 e4) running on the four core 3.4 GHz Intel i7 CPU; for the first time in the run, not all four cores are in use.

The remaining 29 draft 11 positions are:

1 d4 f6/f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e3 f6/f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e4 f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6

[sje]

After 58 weeks:

There are now 373 draft 11 records, 93.25% of the 400 needed. The average draft 11 record generation time is about 26.1 hours.
http://dl.dropbox.com/u/31633927/Perft/Perft13/draft11

There are now 17 draft 12 records, 85% of the 20 needed. The average draft 12 record generation time is about 23.9 days.
http://dl.dropbox.com/u/31633927/Perft/Perft13/draft12

The remaining 27 draft 11 positions are:

1 d4 f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e3 f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e4 f5/g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6

Recent output:

Code: Select all

rnbqkbnr/ppppp1pp/8/8/4Pp2/3B4/PPPPKPPP/RNBQ2NR b kq - 1 3 8 116071939414
rnbqkbnr/ppppp2p/8/5pp1/7Q/4P3/PPPP1PPP/RNB1KBNR b KQkq - 1 3 8 564933423394
rnbqkbnr/ppppp1pp/8/8/4Pp2/3B4/PPPPNPPP/RNBQK2R b KQkq - 1 3 8 172662043234
r1bqkbnr/ppppp1pp/n7/5pB1/3PP3/8/PPP2PPP/RN1QKBNR b KQkq - 0 3 8 741385627909
rnbqkbnr/ppppp1pp/8/8/4PpP1/3B4/PPPP1P1P/RNBQK1NR b KQkq g3 0 3 8 141465399833
r1bqkbnr/ppppp1pp/n7/5pB1/3P1P2/8/PPP1P1PP/RN1QKBNR b KQkq - 0 3 8 168398492365
rnbqkbnr/ppppp2p/8/5pQ1/8/4P3/PPPP1PPP/RNB1KBNR b KQkq - 0 3 8 490706595859
rnbqkbnr/ppppp2p/8/5ppQ/8/4P3/PPPP1PPP/RNB1KBNR b KQkq - 1 3 8 0
rnbqkbnr/ppppp2p/8/5pp1/6Q1/4P3/PPPP1PPP/RNB1KBNR w KQkq - 0 3 9 22068642491896
rnbqkbnr/ppppp1pp/8/8/4Pp2/8/PPPPBPPP/RNBQK1NR b KQkq - 1 3 8 167125289186

[sje]

If each of the ca. 2.0e+18 perft(13) move paths were represented by a half meter chessboard, then that set of chessboards arranged linearly would extend for about 106 light years.

For a one year calculation with electronic charge moving at close to the speed of light, each chess position would VERY ROUGHLY correspond to a chess board about one half centimeter wide. Not surprisingly, this is on the same order as the size of a CPU core.

-----

376 draft 11 records finished, 24 to go:

1 d4 g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e3 g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e4 g6/g5/h6/h5/Na6/Nc6/Nf6/Nh6

[sje]

Of the 400 draft 11 calculations, 379 (94.75%) have been completed; only 21 (5.25%) remain:

1 d4 g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e3 g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e4 g5/h6/h5/Na6/Nc6/Nf6/Nh6

Which will be the very last draft 11 calculation? My guess is 1 e4 Nh6. We should see all of the results in a week or so.

To do the above draft 11 positions, about 190 draft 10 calculations and about 1,200 draft 9 calculations remain.

[sje]

After 59 weeks:

There are now 379 draft 11 records, 94.75% of the 400 needed. The average draft 11 record generation time is about 26.2 hours.
http://dl.dropbox.com/u/31633927/Perft/Perft13/draft11

There are now 17 draft 12 records, 85% of the 20 needed. The average draft 12 record generation time is about 24.3 days.
http://dl.dropbox.com/u/31633927/Perft/Perft13/draft12

About two days of work were lost due to a long, sad attempt to get the wireless card running.

The remaining 21 draft 11 positions are:

1 d4 g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e3 g5/h6/h5/Na6/Nc6/Nf6/Nh6
1 e4 g5/h6/h5/Na6/Nc6/Nf6/Nh6

Recent output:

Code: Select all

rnbqk1nr/ppppppbp/8/3B2p1/8/4P3/PPPP1PPP/RNBQK1NR b KQkq - 3 3 8 599978013643
rnbqk1nr/ppppppbp/4B3/6p1/8/4P3/PPPP1PPP/RNBQK1NR b KQkq - 3 3 8 569614177766
rnbqkb1r/pppppp1p/5n2/6p1/3P4/3BP3/PPP2PPP/RNBQK1NR b KQkq - 2 3 8 763940686495
rnbqk1nr/pppppBbp/8/6p1/8/4P3/PPPP1PPP/RNBQK1NR b KQkq - 0 3 8 36581000009
rnbqk1nr/ppppppbp/8/6p1/2B5/4P3/PPPP1PPP/RNBQK1NR w KQkq - 2 3 9 22625764515797
rnbqkbnr/pppppp2/8/6pp/2B1P3/8/PPPPQPPP/RNB1K1NR b KQkq - 1 3 8 532289213946
rnbqkb1r/pppppp1p/5n2/6p1/2B5/4P3/PPPPQPPP/RNB1K1NR b KQkq - 3 3 8 507387073824
rnbqkb1r/pppppp1p/5n2/6p1/2BP4/4P3/PPP2PPP/RNBQK1NR b KQkq - 2 3 8 715666852061
rnbqkb1r/pppppp1p/5n2/6p1/2B5/4P3/PPPP1PPP/RNBQ1KNR b kq - 3 3 8 469460552111
rnbqkb1r/pppppp1p/5n2/1B4p1/3P4/4P3/PPP2PPP/RNBQK1NR b KQkq - 2 3 8 483455677850

[sje]

Of the 400 draft 11 calculations, 382 (95.5%) have been completed; only 18 (4.5%) remain:

1 d4 h6/h5/Na6/Nc6/Nf6/Nh6
1 e3 h6/h5/Na6/Nc6/Nf6/Nh6
1 e4 h6/h5/Na6/Nc6/Nf6/Nh6

To do the above draft 11 positions, about 160 draft 10 calculations and about 880 draft 9 calculations remain.

Perft(11) for 1 e4 Nh6 is running on a separate machine and with a 120 bit hash. It should be a useful check on the main run.

[sje]

Of the 400 draft 11 calculations, 385 (96.25%) have been completed; only 15 (3.75%) remain:

1 d4 h5/Na6/Nc6/Nf6/Nh6
1 e3 h5/Na6/Nc6/Nf6/Nh6
1 e4 h5/Na6/Nc6/Nf6/Nh6

To do the above draft 11 positions, about 128 draft 10 calculations and about 692 draft 9 calculations remain.