Posts by TJM

1) Message boards : Number crunching : JBIYH paused (Message 13451)
Posted 18 Oct 2017 by Profile TJM
Post:
Most likely it is just a false positive, so the batch is resumed.
2) Message boards : Number crunching : Maximum number of OpenCL GPU tasks (Message 13450)
Posted 18 Oct 2017 by Profile TJM
Post:
Does it work with other projects ?
I've checked your results, it looks like the manager is using only devices #0 - #3 while not running any work on #4 and #5.
3) Message boards : Number crunching : Maximum number of OpenCL GPU tasks (Message 13443)
Posted 17 Oct 2017 by Profile TJM
Post:
Nope, there is no limit, unless you have changed the setting in your project prefs, as there is an option to limit number of tasks.
4) Message boards : Number crunching : JBIYH paused (Message 13409)
Posted 16 Oct 2017 by Profile TJM
Post:
I have paused the JBIYH batches due to an ongoing investigation of possible partial decrypt (below). This also means no tasks for CPU app until the batch is resumed or another one is started, which will take some time.


m3_jbiyh55_2 (JBIYH 55 AV dict set: AVv1 mode 2 n 1)
Hillclimb score: 26774 (00trigr.AVv1)
Machine settings: B:MG:QQI:312:BVCNDYEWFKGXHOJLMPQS
restarts: 193958
Overall score: 30

siegfriqenzwegerensqnixdivxmorvxkigqumiernurdforstngumi
Word search:
Score: 16 (wege sieg nur for nix div )
GC search:
tonixdivx at position 18, match: 0.8
siegfried at position 0, match: 0.8


The first 27 letters are partial match to SIEGFRIEDSIEGFRIEDTONIXDIVX.
This could be a partial decrypt of garbled message or just pure randomness generated by hillclimb.
5) Message boards : Number crunching : Top scores event log (Message 13408)
Posted 15 Oct 2017 by Profile TJM
Post:
I have modified the server code to store all results that have been marked for review by internal filters in a results table(s) that can be viewed here:

Log view for JBIYH
Log view for WEUWY.1 (1124)

It is auto updated as the results are returned if the server has enough spare resources, otherwise it is updated with delay by 3rd server. The overall score listed there ignores the hillclimb score completely, all results that have unique machine settings are passed through algorithms looking for readable words, known signatures, "cribs" and repeated texts. While it does not replace the need for manual review, it might help to find something that looks almost unreadable.

For the batches that are currently running the log above only contains the most recent hits, as processing the full list of results requires lots of processing power. I might do that for some of the batches in the next few days.
6) Message boards : Number crunching : LXACA & JBIYH results for review (Message 13257)
Posted 8 Oct 2017 by Profile TJM
Post:
Alex sent me a link to the mentioned program/algorithm and eventually it is going to be implemented on the server network for realtime/offline result analysis. The readability score next to the results posted above still uses the old algorithm.
7) Message boards : Number crunching : A GPU version of Enigma software (Message 13256)
Posted 8 Oct 2017 by Profile TJM
Post:
Do you have the visual c++ redistributable installed ?
It is the most common thing that causes all the tasks to fail with error 195.
8) Message boards : Number crunching : Not getting GPU work? (Message 13250)
Posted 8 Oct 2017 by Profile TJM
Post:
The server went offline for nearly 24 hours because strong winds have damaged power lines in my region. While it didn't affect the place where I live directly, later they had to cut the power during repairs.
9) Message boards : Bug reports : New AMD driver (Message 13237)
Posted 4 Oct 2017 by Profile TJM
Post:
I encountered some problems while testing on new open-source drivers (Mesa3D 17.0.0 and LLVM 4.0) on my new hardware (RX 480). I need time to troubleshooting. Luckily, main assembly code works very well on opensource drivers (without errors). Only some OpenCL code needs revision.


I'll be releasing matszpk's latest app version under the test platform later today as Enigma GPU test v1.08.
10) Message boards : Bug reports : New AMD driver (Message 13236)
Posted 4 Oct 2017 by Profile TJM
Post:
The Progress indicator does not update for about the first 50 seconds or so the task is running and when it does, the task is then at about 25% complete and the Progress indicator indicates normally until the task ends.


The progress reporting is not currently supported by the GPU apps as they run via wrapper.
The BOINC core client guesses the runtime after a few WUs and then the default progress info seems to fairly accurate.
11) Message boards : News : Enigma Cuda 1.07 (Message 13235)
Posted 4 Oct 2017 by Profile TJM
Post:
The app has it's own checkpoints but since it has no API, the boinc core does not know about that.
12) Message boards : Number crunching : LXACA & JBIYH results for review (Message 13224)
Posted 3 Oct 2017 by Profile TJM
Post:
EDIT: I have updated the post with a dump of nearly 290k unique results for JBIYH at 130,000 full keyspace walks.

We have collected almost 1M of unique results for LXACA.
This message is "special" because it is very short and hillclimb alone probably won't find the solution, there are lots of results with high scores and the actual solution might be hidden somewhere between them.
I passed the results found so far through a filter which tries to estimate the message readability. The files below contain all the unique results and also search results for some of the words that frequently appear in short messages, I'll update the list with more files later.

The format is: decoded text, readability score, hillclimb score

What we are looking for, is text that contains only full words with no garbles, or words separated by "x".

LXACA all results - over 20MB text file

LXACA MORGEN
LXACA TAGE
LXACA MELDUNG
LXACA FRAGE
LXACA FUNKE

JBIYH, format is the same as above: text, readability score (old algorithm), hillclimb score.

JBIYH all unique results
13) Message boards : Number crunching : Future about E@H? (Message 6617)
Posted 29 Sep 2017 by Profile TJM
Post:
The list of unbroken messages is getting shorter, but since we are dealing with the difficult ones, I guess the project will be running for a while.
14) Message boards : Bug reports : New AMD driver (Message 6608)
Posted 24 Sep 2017 by Profile TJM
Post:
Matszpk (the OpenCL port author) said he will look at the possibility to support open source driver.
The current version is highly optimized for AMD OpenCL and that's why it's refusing to run on other platforms.
15) Message boards : Number crunching : A GPU version of Enigma software (Message 6517)
Posted 19 Sep 2017 by Profile TJM
Post:
The current OpenCL version is highly optimized for AMD cards and won't work on Intel GPUs for sure.
Older version did work but it was painfully slow and I've encountered lots of random driver/app crashes, it crashed so often that I could not even verify if it returned correct results, so the support was dropped.
16) Message boards : News : FTNBK broken (Message 6439)
Posted 15 Sep 2017 by Profile TJM
Post:
The 101 letters long FTNBK message was broken.
Please abort all m3_ftnbk* / g3_ftnbk* workunits as they're not needed anymore.

There might be a short break in WU availability as I have to clean up batches that are not needed anymore and prepare new ones.

More info will be posted here: http://www.enigmaathome.net/forum_thread.php?id=1109
17) Message boards : Number crunching : FTNBK broken (Message 6438)
Posted 15 Sep 2017 by Profile TJM
Post:
We have found a partial decrypt of 101 letters long FTNBK message (listed as previously unbroken here: http://cryptocellar.org/bgac/GArmy_July1941.html)

There were a couple of more or less garbled partial decrypts received by the server but it was not noticed until a better solution was found, which is:

Key: B:314:HV:FQR
Plugs: AHCNDFEIKYMPOZRUSWVX
Text: standortderlndxlnkxistxkusownrussixseqsxkvxsuedweztziqxsaitskaxsagoskaxkeineapsfaellexmawhiatvmathiat

Replacing A with O in the 19th group MJKAZ->MJKOZ (http://cryptocellar.org/bgac/16071941_214.pdf ) changes the signature to mathiatvmathiat. It's possible that the plugboard connections are not 100% correct because the message is still garbled, but so far I was unable to find better solution.

I have sent info to Frode Weierud who has already confirmed that the message is broken and I hope that he will post more info soon :D
18) Message boards : Bug reports : Result invalid but credit granted (Message 6392)
Posted 11 Sep 2017 by Profile TJM
Post:
For the test app the first stage of validation was set to accept anything that was processed because we needed the debug data and some of the errors were caused by the application and it's setup.
Later it was changed to filter out errors that were not caused by the app. The host you have linked AFAIR had two AMD cards, one probably damaged or overclocked.
19) Message boards : News : New app: "Enigma Test", OpenCL and Linux tests (Message 6357)
Posted 5 Sep 2017 by Profile TJM
Post:
Cuda app for 64 bit Linux is also available for testing.
1.06 cuda test app uses the same code as 1.10 main app.
1.05 opencl/amd uses matszpk's optimized asm code, once the test are done it'll replace the standard app.
20) Message boards : News : New app: "Enigma Test", OpenCL and Linux tests (Message 6354)
Posted 4 Sep 2017 by Profile TJM
Post:
I have just released 1.05 AMD OpenCL GPU test app for 64bit Linux platform.


Next 20



Copyright © 2017 TJM