Perft(14) Weekly Status Report

[xmas79]

...The MD5 checksum of a file is dependent only upon the contents of the file and not upon the host operating system or the host CPU. Using secure checksums makes it easy to compare file contents for equality without having to have both versions of a file on the same machine.

For the purists:

http://en.wikipedia.org/wiki/Birthday_attack

[syzygy]

...The MD5 checksum of a file is dependent only upon the contents of the file and not upon the host operating system or the host CPU. Using secure checksums makes it easy to compare file contents for equality without having to have both versions of a file on the same machine.

For the purists:

http://en.wikipedia.org/wiki/Birthday_attack

Did you read the whole page?

For comparison, 10^−18 to 10^−15 is the uncorrectable bit error rate of a typical hard disk [1]. In theory, MD5 hashes or UUIDs, being 128 bits, should stay within that range until about 820 billion documents, even if its possible outputs are many more.

[xmas79]

...The MD5 checksum of a file is dependent only upon the contents of the file and not upon the host operating system or the host CPU. Using secure checksums makes it easy to compare file contents for equality without having to have both versions of a file on the same machine.

For the purists:

http://en.wikipedia.org/wiki/Birthday_attack

Blah blah blah blah blah?

Blah blah blah blah blah blah blah.

Here it comes the purist!

[syzygy]

...The MD5 checksum of a file is dependent only upon the contents of the file and not upon the host operating system or the host CPU. Using secure checksums makes it easy to compare file contents for equality without having to have both versions of a file on the same machine.

For the purists:

http://en.wikipedia.org/wiki/Birthday_attack

Blah blah blah blah blah?

Blah blah blah blah blah blah blah.

Here it comes the purist!

Did I step on something?

Feel free to explain your point.

[sje]

I modified my GPU perft program to parse and process the perft(7) records created by Steven for perft(14) calculation.

Tested with work units 0, 33, 36 and 964.

Code: Select all

MD5 (wu7.033.op) = d5a2f1148ef4959bf82fc214e9ebe55a
MD5 (wu7.036.op) = ed92986f663870337811fdb3cf71ab90

036 should finish in another day or so; 033 will need a week or two.

Code: Select all

WU#  Comp%  Machine
---  -----  -------
033  55.4%  megan
036  92.2%  gail

When I get the OpenCL version of Oscar running and I get a high-end video card, we'll likely see the completion of the perft(14) project by the second half of 2015.

But it could take longer, as I avoid all overclocking when I must have accurate results. Consumer quality machines are stressed enough already when running continuously for long periods; the other day I had a 2009 Mac Mini crash and turned off for unknown reasons. When I got it running again, it had somehow permanently disconnected itself from all its audio hardware.

While the hardware failures you see are bad, those which go undetected are far worse.

[sje]

Code: Select all

MD5 (wu7.043.sum) = 36853bc6f7ed0fe03df99380a74629ac

I'll probably start a new document, shared via Dropbox, containing the MD5 signatures of all completed units.

[sje]

MD5 signatures of uncompressed result files are now online at:

https://dl.dropboxusercontent.com/u/31633927/signatures

The above a text file with one line per result file. The file will be updated as the results are produced.

This is a resource for those contributing verification effort to the project. The idea is to allow a quick check of a result file (typical length: ca. 9 MB) without the need to move a lot of data.

[sje]

The calculation of work unit 036 has completed and it matches Ankan's result.

Code: Select all

MD5 (wu7.036.sum) = ed92986f663870337811fdb3cf71ab90

Calculation of work unit 033 continues.

[sje]

Perft(14) Weekly Status 2014-12-14

There are more than 4,100,000 perft(7) results so far, about 4.25% of the 96,400,068 needed.

Day count: 128
Average throughput: 32,031 results/day

Completed work units (42): 000-032, 034-037, 039-040, 042-043, 964
In progress (10): 033, 038, 041, 044-050
Not yet started (913): 051-963

Work units in progress:

Code: Select all

WU#  Comp%  Machine
---  -----  -------
033  61.9%  megan
038  45.8%  joni
041  20.3%  melissa
044  65.3%  cynthia
045  23.2%  betty
046  83.8%  kristen
047  88.4%  amanda
048  26.7%  rocky
049   9.1%  serra
050   1.3%  gail

[ankan]

My weekly status:

work units 964, 0, 33, 36, 38, 41 and 42 completed. 964, 0, 36 and 42 match Steven's results. Other work units can be verified once Steven's machines finish processing them.

I have started with work unit 400 onwards so that some amount of work can be split.
Finished work units 400-425.
426-450 in progress.

Here are md5 checksum of my results (generated using microsoft fciv tool on windows: http://www.microsoft.com/en-in/download ... x?id=11533):

Code: Select all

ee4d4b9dfcaa83d3ca0466e69428596d results\wu7.000.op
d5a2f1148ef4959bf82fc214e9ebe55a results\wu7.033.op
ed92986f663870337811fdb3cf71ab90 results\wu7.036.op
e95fed31de36633f41394621db02cabd results\wu7.038.op
dd1e081f8cf290fd860ec86381d6cb21 results\wu7.041.op
9e396e46c77b2913697506346c4d29eb results\wu7.042.op
ff91186152e5eb3a03edd614b76f2e66 results\wu7.400.op
a1c6f83ca7804832381046585b55bcb9 results\wu7.401.op
4937594aacbbeaa0c1f9074ea5535dd6 results\wu7.402.op
b7365290428b5043507271337c36e4e5 results\wu7.403.op
dd83c73f9173a74a8d5d8dcd765f5f35 results\wu7.404.op
1f8272790b1432cd949c5b690132c857 results\wu7.405.op
511c41e7aec7e7cfa0d888886ee523b4 results\wu7.406.op
ec334f9410323ae7d4c7cba33370e494 results\wu7.407.op
aa3f3fd46ec843262fea5dd2d06801b6 results\wu7.408.op
98e2f2f72a4db1218606420a4d925e9e results\wu7.409.op
9463bba6bc03141cc6f0a449bcdf4fed results\wu7.410.op
5098209ff78410440087cc27f3d5db44 results\wu7.411.op
8f6ba07fe24dd60ec1d8d0f4875c35fe results\wu7.412.op
7e50e2014382d5214e20c0eff166f93f results\wu7.413.op
dc4bc6e95cf327a7290a006719f05803 results\wu7.414.op
a32050d70911882134115176f11413bf results\wu7.415.op
2d7e89941171f2bf028e1f2d1a2eef5a results\wu7.416.op
e89722a099a8ae73fd8106cef084a5f0 results\wu7.417.op
2691bb1e35763867034b0e9e4ff0622a results\wu7.418.op
0c9c9182024b24c497c00c3145f93908 results\wu7.419.op
8148be1fdcb1a35f297450ce65538038 results\wu7.420.op
5a4b44a0c5bcc17984e834643c03de77 results\wu7.421.op
1c3f2b78ec156038d4ee2165ff3bb0f1 results\wu7.422.op
1efd90b34e1f0f63eb649eefe8d69619 results\wu7.423.op
393633645a02766c96b1889f2c8f58fd results\wu7.424.op
4da32c49cd4e9872ac3aad3c320e8f46 results\wu7.425.op
1b329dcb65c93460e614a88823854578 results\wu7.964.op

This is also available at: https://drive.google.com/file/d/0B4d0eT ... sp=sharing

The results are available at:
records 964, 0, 33, 36, 38, 41 and 42: https://drive.google.com/file/d/0B4d0eT ... sp=sharing
records 420-425: https://drive.google.com/file/d/0B4d0eT ... sp=sharing

I also modified my CPU perft program to parse and process work units in parallel and it averages around 1.8 seconds per record (on 4790k processor running 7 threads averaged over first 30,000 records of wu 421). This is almost an order of magnitude slower than my GPU perft program running on GTX 970.