GPU app

Message boards : News : GPU app

To post messages, you must log in.

AuthorMessage
Profile TJM
Project administrator
Project developer
Project scientist
Avatar

Send message
Joined: 25 Aug 07
Posts: 843
Credit: 267,994,998
RAC: 0
Message 5298 - Posted: 26 Jun 2017, 18:22:48 UTC
Last modified: 26 Jun 2017, 18:25:45 UTC

As some of you may have already noticed, there is a GPU version of Enigma software available for testing. It requires nvidia GPU, and probably will run on anything that supports CUDA, I have tested it myself on a couple of low end cards and it ran just fine, except that the system was lagging and WU runtime was very long.

Until yesterday it required app_info to run, this was changed and currently it will be sent to any host that has 'beta work' allowed in preferences. I have reset the beta flag to off for everyone.

The app is marked as "beta" because there are still some issues both on the client and server side. Also, currently the GPU app runs via wrapper and that causes some issues, for example, this page says that interrupting GPU app while it is running the kerney may cause system crash. This is worst case scenario and so far I haven't seen it myself, but I assume it's possible.

Other than that, if left untouched, the app runs smoothly. At this moment it'll only run on CUDA device #0.

Progress bar does not work with the current version of the wrapper, but it's not a high priority thing at this moment, as tasks are relativery short: from ~20 minutes on GTX1050 to around 3m30s on 1080Ti. The app uses internal checkpoints and will automatically resume from the last checkpoint if restarted.
M4 Project homepage
M4 Project wiki
ID: 5298 · 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 5303 - Posted: 26 Jun 2017, 22:03:34 UTC - in response to Message 5298.  

Please take a look at this forum thread before enabling GPU/beta work.
M4 Project homepage
M4 Project wiki
ID: 5303 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
10esseeTony

Send message
Joined: 4 Nov 14
Posts: 3
Credit: 39,379,965
RAC: 0
Message 5357 - Posted: 3 Jul 2017, 17:34:07 UTC

Good work!

Any thoughts on when I can run this on CUDA devices #1, #2, and #3?

Other info that may assist developement:
Currently only device #0 is being utilized, but there are 4 instances running in BOINC Manager. In actuality, the first card is running all 4 tasks simultaneously, but the manager is reporting each task is on a different GPU. (hostid=215904)
ID: 5357 · 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 5366 - Posted: 4 Jul 2017, 7:33:34 UTC - in response to Message 5357.  
Last modified: 4 Jul 2017, 7:34:07 UTC

I'll ask the author to add support for "--device #" passed via command line.

Also, I have removed x86 and x86_64 plan classes "cuda" as the app requires compute capability 2.0+ anyway. Probably after next app upgrade I'll leave just the "cuda_fermi" plan class.
M4 Project homepage
M4 Project wiki
ID: 5366 · 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 5803 - Posted: 8 Jul 2017, 16:15:59 UTC

It would be difficult to add a "long" option, like "--device N", to enigma-cuda. Would a short option, "-d N", solve the problem?
ID: 5803 · 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 5805 - Posted: 8 Jul 2017, 17:50:19 UTC - in response to Message 5803.  
Last modified: 8 Jul 2017, 18:31:44 UTC

Maybe this post can be of help? The dev on xansons fixed the exact same problem.

http://xansons4cod.com/xansons4cod/forum_thread.php?id=41&postid=208#208
ID: 5805 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Dzordzik

Send message
Joined: 11 Mar 17
Posts: 13
Credit: 46,891,270
RAC: 0
Message 5814 - Posted: 9 Jul 2017, 7:04:56 UTC - in response to Message 5303.  

Hi, before some days I found info about Enigma on GPU. I have 2x 1080Ti on my rig but when I install Visual Studio 2015 c++ redistributable package (32 and 64 bit too), add rows do cc and app config and enable running beta and GPU in Enigma settings, all my workunits end after first few seconds of computig with error.
You can look on it and help me how solve it?
http://www.enigmaathome.net/results.php?userid=145166

Errors is same on all units, firs was on Enigma Cuda v1.06 (cuda_fermi) windows_x86_64, last was on Enigma Cuda v1.07 (cuda_fermi) windows_x86_64.

Error: 195 (0x000000C3) EXIT_CHILD_FAILED

I add to cc config:
<exclude_gpu>
<url>http://www.enigmaathome.net/</url>
<device_num>1</device_num>
</exclude_gpu>

to exclude computing on my second GPU

and add to app config:
<app_config>
<app>
<name>enigma_cuda</name>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>

Still have problems and I cannot compute Enigma on my GPU.

<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
13:29:02 (8472): wrapper: running enigma_cuda.exe (-R -o results.txt 04trigr.cur 04unigr.cur 04ciphertext)
app exit status: 0xc000001d
13:29:05 (8472): called boinc_finish

</stderr_txt>
]]>

Can someone help me please?
ID: 5814 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Dzordzik

Send message
Joined: 11 Mar 17
Posts: 13
Credit: 46,891,270
RAC: 0
Message 5816 - Posted: 9 Jul 2017, 7:13:08 UTC - in response to Message 5814.  

And I use last nvidia drivers.
ID: 5816 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
[SG]Felix

Send message
Joined: 1 Oct 15
Posts: 4
Credit: 2,904,215
RAC: 0
Message 13512 - Posted: 3 Nov 2017, 9:41:48 UTC
Last modified: 3 Nov 2017, 9:43:32 UTC

i dont get workunits at all on host 215852

can you look in the server log? because i dont get an scheduler request completed after updating the project

edit: problem solved
ID: 13512 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
jay

Send message
Joined: 26 Nov 17
Posts: 1
Credit: 47,308,546
RAC: 0
Message 13555 - Posted: 29 Nov 2017, 3:31:25 UTC - in response to Message 13512.  

I have to manually request work units, any way around that?
ID: 13555 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Aurel

Send message
Joined: 26 Sep 12
Posts: 18
Credit: 921,616
RAC: 0
Message 13560 - Posted: 2 Dec 2017, 20:00:15 UTC - in response to Message 13555.  
Last modified: 2 Dec 2017, 20:01:05 UTC

I have to manually request work units, any way around that?


Check your BOINC network settings.
BOINC manager->Options->Computing preferences->Network.

Maybe one of those options under "Other" is setted, e.g. "Confirm before connecting to network".
That should do the trick.

Oh, I still dont get any work for GPU. All settings are okay, on other projects I´m getting wu´s.
ID: 13560 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : News : GPU app




Copyright © 2024 TJM