SETI@HOME will only give cpu work units or gpu work units but not both.

Questions and Answers : Unix/Linux : SETI@HOME will only give cpu work units or gpu work units but not both.
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile William E. Strong

Send message
Joined: 11 Sep 99
Posts: 4
Credit: 84,869,944
RAC: 74
United States
Message 2016340 - Posted: 23 Oct 2019, 5:55:58 UTC

I have connected to seti project 3 times. My system is capable of cpu and gpu work units. First time I connected the project sent cpu work units, and if you checked project property's the gpu was being kept in infinite time out.
The second time I connected I had gpu work units! and no cpu work units, and if you checked project property's the cpu was in time out! The third time I connected I got cpu work units and if you checked project property's
it says "don't ask for gpu work units project doesn't have any".

How do I get both cpu and gpu work units?
ID: 2016340 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13063
Credit: 208,696,464
RAC: 304
Australia
Message 2016342 - Posted: 23 Oct 2019, 6:04:05 UTC

Which system are you referring to?
I just had a quick look at all of the active systems on your account, and all of them (that have a GPU) have both CPU & GPU Work allocated for them to process.
Grant
Darwin NT
ID: 2016342 · Report as offensive
Profile William E. Strong

Send message
Joined: 11 Sep 99
Posts: 4
Credit: 84,869,944
RAC: 74
United States
Message 2016380 - Posted: 23 Oct 2019, 15:41:49 UTC - in response to Message 2016342.  

The system I refer to is sixcore (linux) not windows, amd threadripper 12 core cpu 2 evga 1070 gpu's
ID: 2016380 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 18997
Credit: 416,307,556
RAC: 380
United Kingdom
Message 2016401 - Posted: 23 Oct 2019, 19:32:54 UTC

It would appear you are using a member of the 436.x series of GPU drivers - these are known to give issues. Try rolling back to a member of the 430.x series.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 2016401 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 11914
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2016410 - Posted: 23 Oct 2019, 21:36:25 UTC

It doesn't matter about the Nvidia driver version in Linux. All the recent versions are usable and have no issues. I assume you are referring to host https://setiathome.berkeley.edu/show_host_detail.php?hostid=8830389

I don't see normal anonymous platform gpu apps shown in the Tasks in Progress. All I see are some CUDA60 tasks that probably were downloaded at first connect on the stock BOINC installation. I also see 30 abandoned CUDA60 tasks. If you are running an anonymous platform you have to define which apps you want to run. The older CUDA60 app is a poor choice for a 1070 and you would have much better crunching if you used the standard SoG gpu application. Normally when running the standard BOINC platform, the scheduler tries out all varieties of applications to determine which has the best performance on your hardware. It normally figures out the SoG app is the best performer after a couple of weeks.

Be aware that to run the SoG application which is a OpenCL application that you also need to have the OpenCL component of the Nvidia drivers installed to run the app. Sometimes, that component does not get installed and you have to manually install it on Linux/Debian.

Also since you are running Linux and modern Nvidia gpu hardware, there is another option and that is to run the TBar BOINC All-in-One installer and that installs BOINC into the /home folder for easy access and also installs the most optimized cpu and gpu applications. The gpu applications are either CUDA92 or CUDA101 and are 3X-5X faster than the stock gpu applications.

You can read about the AIO installer here:
https://setiathome.berkeley.edu/forum_thread.php?id=81271
And you can download the installer here:
http://www.arkayn.us/lunatics/BOINC.7z
Just unpack the archive somewhere convenient in /home and read the readme. If you have dependencies met then just double-clicking the boincmgr file starts the client and Manager and presents the normal Join Project dialogue. All the applications are already installed and configured in the anonymous platform app_info.xml file. Soon as you join Seti, the client will contact the scheduler and download both cpu and gpu work for your host and start crunching both.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2016410 · Report as offensive
Profile William E. Strong

Send message
Joined: 11 Sep 99
Posts: 4
Credit: 84,869,944
RAC: 74
United States
Message 2016455 - Posted: 24 Oct 2019, 5:19:23 UTC - in response to Message 2016410.  

Yes that's the computer and that's it's history. Looks like I have much to learn and much to do. I will report back. Thank You!
ID: 2016455 · Report as offensive
Profile William E. Strong

Send message
Joined: 11 Sep 99
Posts: 4
Credit: 84,869,944
RAC: 74
United States
Message 2016895 - Posted: 27 Oct 2019, 21:28:04 UTC

I don't know where the thread that got started around my question "boinc will do pc work units or gpu work units but not both" went, but it fixed my problem. I downloaded the version of boinc you listed installed it (but with an anonymous xml.) I then read up on xml and then how boinc used xml. That's were I found out about anonymous xml's that you told me about. I purged my client of work units and removed app_xml reset the project. Please reply I would love to know how to keep up on the latest versions of boinc.
















0
ID: 2016895 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 11914
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2017027 - Posted: 28 Oct 2019, 20:38:44 UTC
Last modified: 28 Oct 2019, 20:54:37 UTC

It looks like you either reinstalled the AIO version of BOINC or installed for the first time and the special app just recently. I see nothing but the stock CUDA60 gpu tasks in your list and the first of the AIO special app tasks. Only 1 task in progress though because of all the abandoned CUDA60 tasks. You are in the penalty box right now until the latest special app task gets validated. For each new validated task, Seti will send you two more until you build back up to your normal gpu cache level. Your first special app task is here:
https://setiathome.berkeley.edu/result.php?resultid=8178056988
It completed in 93 seconds on your 1070. Compare that to any of your tasks done by the old CUDA60 app which were taking 1300 seconds and you can see why the special app is recommended on Nvidia hardware in Linux hosts.
Just have patience to get out of the penalty box. You get sent one cpu task and one gpu task per day if you have any errors. Should correct itself fairly fast once your wingmen report in.

[Edit] The host is already pulling itself out of the penalty box. 53 in progress and lots of new special app pendings.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2017027 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 11914
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2017032 - Posted: 28 Oct 2019, 20:49:21 UTC

I would just stand pat for a while with your current stock AIO app_info file until you get out of the penalty box and get back to a normal cache level. Later, I would suggest adding the -nobs parameter to either the app_info or a app_config file command line entry in those files. That makes the cpu support the gpu task with a full cpu thread and speeds up the calculation by 5-15 seconds normally and you will see that the cpu_time will equal the run_time for the task. Read the README_x41p_V0.98.txt in the /BOINC/projects/setiathome.berkeley.edu/docs/docs directory for explanation of where to add the command.

5) If you wish to use 100% CPU per task, add the command -nobs to the CUDA section of the app_info.xml,
<plan_class>cuda90</plan_class>
<cmdline>-nobs</cmdline>
<coproc>
There isn't any requirement to use a Full CPU per task, but, it may be a few seconds faster.

You have plenty of cpu threads with your 1920X cpu to use -nobs.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2017032 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 11914
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2017041 - Posted: 28 Oct 2019, 23:11:13 UTC

Your gpu task times just took a turn for the worse. Are you trying to run multiple concurrent tasks on the card now? That does not work very well with the special apps. You should only run singles on the gpus.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2017041 · Report as offensive

Questions and Answers : Unix/Linux : SETI@HOME will only give cpu work units or gpu work units but not both.


 
©2020 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.