A GPU version of Enigma software

Message boards : Number crunching : A GPU version of Enigma software

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 . . . 12 · Next

AuthorMessage
Profile Sabroe_SMC

Send message
Joined: 2 Feb 10
Posts: 4
Credit: 39,130,140
RAC: 0
Message 5342 - Posted: 30 Jun 2017, 19:02:10 UTC
Last modified: 30 Jun 2017, 19:02:57 UTC

Crunching since 3 days on a not aced 1070 and i see mistakes. 18 of 140 WUs are exiting with error 195
Is this normal? They are only lasting 2 seconds so that's not a big deal.
ID: 5342 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
JugNut

Send message
Joined: 16 Mar 13
Posts: 24
Credit: 125,506,046
RAC: 0
Message 5344 - Posted: 1 Jul 2017, 12:30:04 UTC - in response to Message 5342.  
Last modified: 1 Jul 2017, 12:31:10 UTC

Hi guy's I just tried a pair of GTX 580's with same result error 195. Am I right is saying the fermi GPU app does NOT support most fermi"s?

This error would seem to support that theory, the STDERR reads.. "Program requires GPU with compute capability 2.1 or higher" Meaning most fermi's will not work here. http://www.enigmaathome.net/result.php?resultid=274421739

Later model fermi's such as the GTX 560 ti might work as they have compute cap. v2.1. Most other fermi including the 560, 570, 580, 590 only have compute capability v2.0 so should not work.

Anyone confirm?
ID: 5344 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 5345 - Posted: 1 Jul 2017, 13:23:39 UTC - in response to Message 5342.  
Last modified: 1 Jul 2017, 13:25:01 UTC

Crunching since 3 days on a not aced 1070 and i see mistakes. 18 of 140 WUs are exiting with error 195
Is this normal? They are only lasting 2 seconds so that's not a big deal.


Must be a bug of some sort, but that's either a bug in the app or on the server side. 195 means that the app could not start so perhaps one of the workunit files is missing or corrupted. If it fails on one host, it will also fail for everyone running a resend of affected WU. Just ignore this for now, the server saves a lot of debug info for workunits which fail, so I should be able to find the reason and possibly fix the problem soon.
M4 Project homepage
M4 Project wiki
ID: 5345 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Alex VE3NEA

Send message
Joined: 22 Oct 16
Posts: 36
Credit: 128,745
RAC: 0
Message 5346 - Posted: 1 Jul 2017, 14:12:32 UTC - in response to Message 5344.  

On start-up, enigma-cuda checks the cuda compute capability of the video card and rejects those cards that are below 2.1. This is not because 2.1 has something that 2.0 doesn't, but because the program is optimized for the modern cards and is very slow on the old ones. On my GTX-420, it runs several times slower than the CPU-based enigma-AV.

I could change the code to allow the 2.0 cards, but this does not make much sense to me as the people with such cards would be more productive if they ran enigma-av instead.

Perhaps it makes more sense to set the threshold at 3.0 so that only the cards with at least the Kepler architecture are allowed to run enigma-cuda.

What are your thoughts?
ID: 5346 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 5347 - Posted: 1 Jul 2017, 15:42:51 UTC - in response to Message 5346.  
Last modified: 1 Jul 2017, 15:43:40 UTC

IMO limiting the cards at 2.1 does not make sense, aren't all the cards listed as 2.0 actually much faster than majority of the compute capability 2.1 cards ?

https://en.wikipedia.org/wiki/CUDA#GPUs_supported
https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units
M4 Project homepage
M4 Project wiki
ID: 5347 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
JugNut

Send message
Joined: 16 Mar 13
Posts: 24
Credit: 125,506,046
RAC: 0
Message 5348 - Posted: 1 Jul 2017, 17:55:14 UTC - in response to Message 5347.  
Last modified: 1 Jul 2017, 18:05:03 UTC

The GTX 580 is still a decent card, it's certainly inefficient power wise but in most projects I find it performs similarly to a GTX 670. Of course that's still not saying much either but hey, I have what I have. I'm just not sure that ruling out users because there GPU's may not be fast enough is right way to go either? Perhaps test first to see what works & what doesn't.

I'd still like to give the 580's a try, but if it turned out as bad as you say then of course i'd put them to work at a different project where they still have value. I do have better cards so either way I wont miss out.

At the end of the day it's for you guys to decide.

All the best..
ID: 5348 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Alex VE3NEA

Send message
Joined: 22 Oct 16
Posts: 36
Credit: 128,745
RAC: 0
Message 5349 - Posted: 2 Jul 2017, 1:31:27 UTC

I have rebuilt enigma-cuda.exe to allow compute capability 2.0, download the new executables from GitHub. Warning: untested, test yourself.
ID: 5349 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 5350 - Posted: 2 Jul 2017, 8:53:10 UTC - in response to Message 5348.  


I'd still like to give the 580's a try, but if it turned out as bad as you say then of course i'd put them to work at a different project where they still have value. I do have better cards so either way I wont miss out.


I have prepared test suite for you to try out:

https://s3.eu-central-1.amazonaws.com/enigma4/test_runs/2.0_test.zip

Inside you'll find 3 bat files to run enigma-cuda with different settings, included executable is 64bit, replace it with 32bit version if you need. Check if it works and tell us what are the runtimes - the app reports runtime once it's done.
M4 Project homepage
M4 Project wiki
ID: 5350 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 5351 - Posted: 2 Jul 2017, 12:25:37 UTC

ID: 5351 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Empie

Send message
Joined: 5 Mar 10
Posts: 12
Credit: 40,797,951
RAC: 0
Message 5352 - Posted: 2 Jul 2017, 14:07:14 UTC - in response to Message 5345.  

Crunching since 3 days on a not aced 1070 and i see mistakes. 18 of 140 WUs are exiting with error 195
Is this normal? They are only lasting 2 seconds so that's not a big deal.


Must be a bug of some sort, but that's either a bug in the app or on the server side. 195 means that the app could not start so perhaps one of the workunit files is missing or corrupted. If it fails on one host, it will also fail for everyone running a resend of affected WU. Just ignore this for now, the server saves a lot of debug info for workunits which fail, so I should be able to find the reason and possibly fix the problem soon.


Same here on a 1070. Swapped my Asus Strix 970 for an EVGA 1070 SC yesterday. I haven't seen errors on the 970, but since I have the 1070, I see those 195 errors too. Takes 3 secs to fail. Resends seem te complete fine. I did update the driver to 384.76 .
Hope this helps.
ID: 5352 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
JugNut

Send message
Joined: 16 Mar 13
Posts: 24
Credit: 125,506,046
RAC: 0
Message 5353 - Posted: 2 Jul 2017, 15:28:54 UTC - in response to Message 5350.  

Thanks kindly Alex & TJM, i'll give this a crack as soon :)
ID: 5353 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
forretrio
Avatar

Send message
Joined: 25 Dec 14
Posts: 9
Credit: 5,541,534
RAC: 0
Message 5354 - Posted: 3 Jul 2017, 8:12:05 UTC

http://www.enigmaathome.net/result.php?resultid=275447829

<core_client_version>7.6.33</core_client_version>
<![CDATA[
<message>
(unknown error) - exit code 195 (0xc3)
</message>
<stderr_txt>
wrapper [26118 718 + noargs]: starting
20:04:45 (18036): wrapper: running enigma_cuda.exe (-R -o results.txt 04trigr.cur 04unigr.cur 04ciphertext)
CUDA error: C:/Proj/Other/Enigma/enigma-cuda/enigma-cuda-lib/cuda_code.cu:798:  CUDA driver version is insufficient for CUDA runtime versionInitialization failed: CUDA error: C:/Proj/Other/Enigma/enigma-cuda/enigma-cuda-lib/cuda_code.cu:798:  CUDA driver version is insufficient for CUDA runtime version
app exit status: 0x2
20:04:48 (18036): called boinc_finish

</stderr_txt>
]]>


Weird...I am using kind of a new laptop afterall.

950M should be working as indicated in the GPU list :d
ID: 5354 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Alex VE3NEA

Send message
Joined: 22 Oct 16
Posts: 36
Credit: 128,745
RAC: 0
Message 5355 - Posted: 3 Jul 2017, 10:55:35 UTC - in response to Message 5354.  
Last modified: 3 Jul 2017, 10:58:06 UTC

@forretrio, it looks like you just need to install the latest driver for your video card.

The drivers are here: http://www.nvidia.com/Download/index.aspx
ID: 5355 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
JugNut

Send message
Joined: 16 Mar 13
Posts: 24
Credit: 125,506,046
RAC: 0
Message 5356 - Posted: 3 Jul 2017, 15:03:19 UTC - in response to Message 5350.  
Last modified: 3 Jul 2017, 15:04:39 UTC

Hi Alex & TJM unfortunately the cap. v2.0 test app didn't fare well on the GTX 580 and crashed immediately on start up for all 3 batch files tested.

I have sent TJM the windows logs in hope something might be gleaned from them.

If you need anything further just ask.

Thanks for the assist.
ID: 5356 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Dave GPU

Send message
Joined: 3 Dec 14
Posts: 10
Credit: 2,223,253,098
RAC: 0
Message 5358 - Posted: 3 Jul 2017, 17:36:38 UTC

I have 4 rigs 2 t3500 Xeon machines. & 2 dual cores, 1 amd & 1 intel, that will not run the app
New install of win10 on the t3500 + VC2015 still no go. Pulled intel 2 core and install ( i5 4 core and it worked.)

How do I make the Xeon work, Thanks.
ID: 5358 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 5359 - Posted: 3 Jul 2017, 18:23:04 UTC - in response to Message 5358.  

Yesterday I have released test M3 batch which has failed.
The app expects lower case UKW & left ring when running M4 task from resume (upper case won't work), but will fail with lower case UKW when running M3 task.
I can easily work around that on the server side, since the input files are autogenerated it's just a minor tweak to use mixed case for M4 and all upper case for M3, but future app versions must be able to read resume files like this - once the batch is running, the input files have their checksums stored in the database and any change will result in download errors.
M4 Project homepage
M4 Project wiki
ID: 5359 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 5360 - Posted: 3 Jul 2017, 18:28:08 UTC - in response to Message 5358.  

I have 4 rigs 2 t3500 Xeon machines. & 2 dual cores, 1 amd & 1 intel, that will not run the app
New install of win10 on the t3500 + VC2015 still no go. Pulled intel 2 core and install ( i5 4 core and it worked.)

How do I make the Xeon work, Thanks.


Run the executable from command line and check for any error messages.
You can use the executable from project folder or just download this https://s3.eu-central-1.amazonaws.com/enigma4/test_runs/2.0_test.zip and run one of the bat files.
M4 Project homepage
M4 Project wiki
ID: 5360 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
mmonnin

Send message
Joined: 2 Jul 16
Posts: 26
Credit: 67,848,645
RAC: 0
Message 5361 - Posted: 3 Jul 2017, 19:00:47 UTC - in response to Message 5354.  
Last modified: 3 Jul 2017, 19:01:11 UTC

http://www.enigmaathome.net/result.php?resultid=275447829

<core_client_version>7.6.33</core_client_version>
<![CDATA[
<message>
(unknown error) - exit code 195 (0xc3)
</message>
<stderr_txt>
wrapper [26118 718 + noargs]: starting
20:04:45 (18036): wrapper: running enigma_cuda.exe (-R -o results.txt 04trigr.cur 04unigr.cur 04ciphertext)
CUDA error: C:/Proj/Other/Enigma/enigma-cuda/enigma-cuda-lib/cuda_code.cu:798:  CUDA driver version is insufficient for CUDA runtime versionInitialization failed: CUDA error: C:/Proj/Other/Enigma/enigma-cuda/enigma-cuda-lib/cuda_code.cu:798:  CUDA driver version is insufficient for CUDA runtime version
app exit status: 0x2
20:04:48 (18036): called boinc_finish

</stderr_txt>
]]>


Weird...I am using kind of a new laptop afterall.

950M should be working as indicated in the GPU list :d


I got some of these today as well. GTX 1070 Win7.
ID: 5361 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Alex VE3NEA

Send message
Joined: 22 Oct 16
Posts: 36
Credit: 128,745
RAC: 0
Message 5363 - Posted: 4 Jul 2017, 0:15:59 UTC

if the error message says that "CUDA driver version is insufficient for CUDA runtime version", all you have to do to fix this is install the latest video driver from the Nvidia web site. This may be necessary even if your laptop is new.
ID: 5363 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
forretrio
Avatar

Send message
Joined: 25 Dec 14
Posts: 9
Credit: 5,541,534
RAC: 0
Message 5364 - Posted: 4 Jul 2017, 4:02:34 UTC

Thanks for the advice. It worked properly for the first few tasks.
ID: 5364 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 . . . 12 · Next

Message boards : Number crunching : A GPU version of Enigma software




Copyright © 2024 TJM