Perft(14) Weekly Status Reports for 2015

[sje]

Perft(14) Weekly Status 2015-02-15

Symbolic has produced more than 6,800,000 perft(7) results so far, about 7.05% of the 96,400,068 needed.

Day count: 191
Average throughput: 35,602 results/day

Completed work units (69): 000-040 042-058 060-062 064-067 069-071 964
Independently verified (6): 000 033 036 038 042 964
In progress (10): 041 059 063 068 072-077
Not yet started (886): 078-963

Work units in progress:

Code: Select all

WU#  Comp%  Machine
---  -----  -------
041   85.1  melissa
059   91.1  megan
063   50.5  joni
068   22.4  betty
072   29.4  gail
073   45.8  kristen
074   16.3  serra
075   34.9  amanda
076    7.4  cynthia
077    8.8  rocky

--------

Independent results from Arkan and his program:

Test work units (7): 000 033 036 038 041-042 964
Production units (200): 400-599

--------

Total distinct completed work units (270): 000-058 060-062 064-067 069-071 400-599 964
Total uncompleted work units (695): 059 063 068 072-399 600-963
Project completion: 27.98%

[sje]

For each work unit completed by Symbolic, the maximum perft(7) calculation is identified and the corresponding result record is copied to the Largest list. This list can be accessed at:
https://dl.dropboxusercontent.com/u/316 ... 14/Largest

Example entry:

Code: Select all

wu7.000.sum
1nbqkbnr/1ppp1ppp/4p3/r7/6Q1/3BP3/PPPP1PPP/RNB1K1NR b KQk - 3 4 1 159206632060 159206632060

There are currently 69 entries on the list; more will be added as the results become available.

The main reason for presenting this list is to give others the opportunity to easily assist with verification. The idea here is that the largest results are somehow most likely to contain errors.

[sje]

Arkan's program has completed work units 400 through 599. I have scanned his result files and have copied out the maximum perft(7) record in each. Here they are:
https://dl.dropboxusercontent.com/u/316 ... rgestArkan

If you see a problem with these or with any of Symbolic's calculations, please post the details.

[sje]

Perft(14) Weekly Status 2015-02-22

Symbolic has produced more than 7,000,000 perft(7) results so far, about 7.26% of the 96,400,068 needed.

Day count: 198
Average throughput: 35,354 results/day

Completed work units (71): 000-040 042-062 064-067 069-071 075 964
Independently verified (6): 000 033 036 038 042 964
In progress (10): 041 063 068 072-074 076-079
Not yet started (884): 080-963

Work units in progress:

Code: Select all

WU#  Comp%  Machine
---  -----  -------
041   91.2  melissa
063   58.2  joni
068   28.1  betty
072   52.4  gail
073   94.2  kristen
074   45.8  serra
076   37.8  cynthia
077   59.4  rocky
078   10.2  megan
079    2.8  amanda

--------

Independent results from Arkan and his program:

Test work units (7): 000 033 036 038 041-042 964
Production units (200): 400-599

--------

Total distinct completed work units (272): 000-062 064-067 069-071 075 400-599 964
Total uncompleted work units (693): 063 068 072-074 076-399 600-963
Project completion: 28.19%

[Henk]

Perft(14) Weekly Status 2015-02-22

Symbolic has produced more than 7,000,000 perft(7) results so far, about 7.26% of the 96,400,068 needed.

Day count: 198
Average throughput: 35,354 results/day

Completed work units (71): 000-040 042-062 064-067 069-071 075 964
Independently verified (6): 000 033 036 038 042 964
In progress (10): 041 063 068 072-074 076-079
Not yet started (884): 080-963

Work units in progress:

Code: Select all

WU#  Comp%  Machine
---  -----  -------
041   91.2  melissa
063   58.2  joni
068   28.1  betty
072   52.4  gail
073   94.2  kristen
074   45.8  serra
076   37.8  cynthia
077   59.4  rocky
078   10.2  megan
079    2.8  amanda

--------

Independent results from Arkan and his program:

Test work units (7): 000 033 036 038 041-042 964
Production units (200): 400-599

--------

Total distinct completed work units (272): 000-062 064-067 069-071 075 400-599 964
Total uncompleted work units (693): 063 068 072-074 076-399 600-963
Project completion: 28.19%

Perhaps after hundred years they compute all this within an hour.

[vittyvirus]

Perft(14) Weekly Status 2015-02-22

Symbolic has produced more than 7,000,000 perft(7) results so far, about 7.26% of the 96,400,068 needed.

Day count: 198
Average throughput: 35,354 results/day

Completed work units (71): 000-040 042-062 064-067 069-071 075 964
Independently verified (6): 000 033 036 038 042 964
In progress (10): 041 063 068 072-074 076-079
Not yet started (884): 080-963

Work units in progress:

Code: Select all

WU#  Comp%  Machine
---  -----  -------
041   91.2  melissa
063   58.2  joni
068   28.1  betty
072   52.4  gail
073   94.2  kristen
074   45.8  serra
076   37.8  cynthia
077   59.4  rocky
078   10.2  megan
079    2.8  amanda

--------

Independent results from Arkan and his program:

Test work units (7): 000 033 036 038 041-042 964
Production units (200): 400-599

--------

Total distinct completed work units (272): 000-062 064-067 069-071 075 400-599 964
Total uncompleted work units (693): 063 068 072-074 076-399 600-963
Project completion: 28.19%

Perhaps after hundred years they compute all this within an hour.

That'd be the time when we will be trying to solve Go, after solving chess.
PS Be it calculating Perft(15) or trillions of digits of pi, it's more a kind of sport.

[Henk]

Perft(14) Weekly Status 2015-02-22

Symbolic has produced more than 7,000,000 perft(7) results so far, about 7.26% of the 96,400,068 needed.

Day count: 198
Average throughput: 35,354 results/day

Completed work units (71): 000-040 042-062 064-067 069-071 075 964
Independently verified (6): 000 033 036 038 042 964
In progress (10): 041 063 068 072-074 076-079
Not yet started (884): 080-963

Work units in progress:

Code: Select all

WU#  Comp%  Machine
---  -----  -------
041   91.2  melissa
063   58.2  joni
068   28.1  betty
072   52.4  gail
073   94.2  kristen
074   45.8  serra
076   37.8  cynthia
077   59.4  rocky
078   10.2  megan
079    2.8  amanda

--------

Independent results from Arkan and his program:

Test work units (7): 000 033 036 038 041-042 964
Production units (200): 400-599

--------

Total distinct completed work units (272): 000-062 064-067 069-071 075 400-599 964
Total uncompleted work units (693): 063 068 072-074 076-399 600-963
Project completion: 28.19%

Perhaps after hundred years they compute all this within an hour.

That'd be the time when we will be trying to solve Go, after solving chess.
PS Be it calculating Perft(15) or trillions of digits of pi, it's more a kind of sport.

Not a very active sport. Start and wait. Only do something when a machine goes down.

[sje]

Perft(14) Weekly Status 2015-03-01

Symbolic has produced more than 7,200,000 perft(7) results so far, about 7.47% of the 96,400,068 needed.

Day count: 205
Average throughput: 35,122 results/day

Completed work units (73): 000-040 042-062 064-067 069-071 073 075 077 964
Independently verified (6): 000 033 036 038 042 964
In progress (10): 041 063 068 072 074 076 078-081
Not yet started (882): 082-963

Work units in progress:

Code: Select all

WU#  Comp%  Machine
---  -----  -------
041   97.8  melissa
063   68.8  joni
068   33.8  betty
072   73.8  gail
074   78.4  serra
076   70.2  cynthia
078   31.3  megan
079   73.0  amanda
080   59.7  kristen
081   25.6  rocky

--------

Independent results from Arkan and his program:

Test work units (7): 000 033 036 038 041-042 964
Production units (200): 400-599

--------

Total distinct completed work units (274): 000-062 064-067 069-071 073 075 077 400-599 964
Total uncompleted work units (691): 063 068 072 074 076 078-399 600-963
Project completion: 28.39%

[sje]

Work unit 041 has completed after a 100 day calculation running on an old (2006) Intel Core 2 Duo notebook. This work unit is the last of the seven test units run by Arkan and his program now confirmed by Symbolic. Each and every of the 600,068 perft(7) calculations from Arkan's test work units matches the corresponding result by Symbolic.

Since I've never seen Arkan's program and he's never seen Symbolic, the total agreement so far is good evidence of the correctness of both programs. Also, it's a testimony to the reliability of the hardware involved.

----

I've not heard from Arkan for a while and I hope he's still interested in the project. Also, I'd like to know if his program uses transposition assistance; and if so, the number of bits in a hash signature.

[ankan]

Sorry for not updating the results of my perft14 verification run. I am working on a chess engine and my PC has been busy running tests for it slowing down the perft verification effort. I will try to devote some machine time to perft runs again.

After Steven pointed out some errors in my results I wrote a small program for doing sanity check on my generated results and it found issues with two more work units:

wu7.476: disk or memory read error?
r1bqkbnr/pppppp1p/8/n2P2p1/8/4B1P1/PPP1PP1P/RN1QKBNR b KQkq - 0 4 24
was read as
r1bqkbnr/pppppp1p/8/n2P2p1/8/4B1P1/PPP1PP1P/RN1QKBNR b KQkq - 0 4 r4

wu7.690: some hardware instability resulting in gpu error causing around ten thousand records getting (same) wrong perft value. I should add some error checks in my gpu program to terminate early if it detects launch failure (otherwise the perft value of previous successful record gets repeatedly written for subsequent records).

I have re-run the above units and they completed without any issue second time.

Updated results are available at:

test WUs : https://drive.google.com/file/d/0B4d0eT ... sp=sharing
WUs 400-499: https://drive.google.com/file/d/0B4d0eT ... sp=sharing
WUs 500-599: https://drive.google.com/file/d/0B4d0eT ... sp=sharing
WUs 600-699: https://drive.google.com/file/d/0B4d0eT ... sp=sharing
WUs 700-799: https://drive.google.com/file/d/0B4d0eT ... sp=sharing

checksums:
https://drive.google.com/file/d/0B4d0eT ... sp=sharing


My gpu perft program uses transposition table with only 64 bit hash keys. The hash table is cleared only after finishing a full work unit.

I know this is risky but as this is only a verification run I didn't modify my program to use bigger hash keys.

I have run perft 12 of startpos many times without any issue but I encountered hash key collisions in perft 13 run. As a single work unit (4-6 hours) is smaller than perft 12 run (~12 hours), I am hoping to not get any collisions.

If there is any discrepancy between my result and anyone else's result, I can quickly re-check the result with a different zobrist key set (which would normally give a different result if there was a collision).