SETI@home v8 beta to begin on Tuesday

Message boards : News : SETI@home v8 beta to begin on Tuesday
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 52 · 53 · 54 · 55 · 56 · 57 · 58 . . . 99 · Next

AuthorMessage
Grant (SSSF)
Volunteer tester

Send message
Joined: 15 Jun 16
Posts: 43
Credit: 1,836,741
RAC: 0
Australia
Message 58849 - Posted: 2 Jul 2016, 8:42:08 UTC - in response to Message 58848.  
Last modified: 2 Jul 2016, 8:47:47 UTC

At least it was quickly resolved, unlike Keith Myers intermittent download issues on main.

EDIT-
From the Release notes for v5.24
- Fixed a memory leak in server side of remote sensor monitoring.
Grant
Darwin NT.
ID: 58849 · Report as offensive
BetelgeuseFive
Volunteer tester

Send message
Joined: 3 Jun 12
Posts: 64
Credit: 2,418,759
RAC: 639
Netherlands
Message 58853 - Posted: 2 Jul 2016, 13:49:02 UTC

Why are the new (8.13/8.14) opencl_nvidia tasks (both SoG and sah) using so much CPU time ? I have specified -use_sleep, but I still seem to need nearly an entire CPU core to feed the GPU.

Examples:

8.13 SoG: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24206772
8.14 SoG: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24217433
8.14 SoG: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24225616
8.14 sah: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24225173

Previous versions did not have this problem:

8.12 SoG: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24146549
8.12 SoG: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=23934991
8.12 sah: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=23868462

I don't like wasting my CPU time just feeding the GPU and by using -use_sleep there were no problems (BOINC would soon select cuda50 over opencl versions), but now I am not so sure.

Tom
ID: 58853 · Report as offensive
Profile Raistmer
Volunteer tester
Avatar

Send message
Joined: 18 Aug 05
Posts: 2423
Credit: 15,878,738
RAC: 0
Russia
Message 58854 - Posted: 2 Jul 2016, 14:42:09 UTC - in response to Message 58853.  

Why are the new (8.13/8.14) opencl_nvidia tasks (both SoG and sah) using so much CPU time ? I have specified -use_sleep, but I still seem to need nearly an entire CPU core to feed the GPU.

Add -v 6 to tuning line. And post few results with this option ON.
News about SETI opt app releases: https://twitter.com/Raistmer
ID: 58854 · Report as offensive
Linux? You're kidding me!!
Volunteer tester
Avatar

Send message
Joined: 10 Mar 12
Posts: 1601
Credit: 11,503,026
RAC: 18,345
Sweden
Message 58855 - Posted: 2 Jul 2016, 18:38:31 UTC
Last modified: 2 Jul 2016, 18:45:18 UTC

Yikes!!, we're in the middle of a perfect Arecibo VLAR storm again. No wonder the tasks all of a sudden takes more than twice as long as before :-)

This is going to screw up the APR again, so Non SoG will be chosen just when the Arecibo VLAR storm is over, and then Non SoG will be chosen as the fastest app, because it never did run any Arecibo VLARs.

Just like the last Arecibo VLAR storm, one has to then manually abort so many Non SoG, so that SoG again can come into play. Not good at all when there is so many WU's of one type being split, and not a good mix.
WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.
ID: 58855 · Report as offensive
Linux? You're kidding me!!
Volunteer tester
Avatar

Send message
Joined: 10 Mar 12
Posts: 1601
Credit: 11,503,026
RAC: 18,345
Sweden
Message 58856 - Posted: 2 Jul 2016, 21:25:31 UTC
Last modified: 2 Jul 2016, 21:31:47 UTC

Dropping on the fly <ngpus>, to 0.50 (from 0.33) for SoG and Non SoG, in app_config.xml, because I already know that SoG is the fastest app, no matter the AR.

Just to avoid APR rot, getting slower apps, due to this Arecibo VLAR storm.
WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.
ID: 58856 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 15 Jun 16
Posts: 43
Credit: 1,836,741
RAC: 0
Australia
Message 58857 - Posted: 2 Jul 2016, 23:25:59 UTC - in response to Message 58856.  

Dropping on the fly <ngpus>, to 0.50 (from 0.33) for SoG and Non SoG, in app_config.xml, because I already know that SoG is the fastest app, no matter the AR.

Just to avoid APR rot, getting slower apps, due to this Arecibo VLAR storm.

That's what I do change applications, no need to abort. Just load up the undesirable applications with 2, 3, or 4 WUs per GPU and watch it's APR plummet, then go back to the usual values when the VLAR or Guppie storm is over and you've got your preferred application back.
Grant
Darwin NT.
ID: 58857 · Report as offensive
Linux? You're kidding me!!
Volunteer tester
Avatar

Send message
Joined: 10 Mar 12
Posts: 1601
Credit: 11,503,026
RAC: 18,345
Sweden
Message 58858 - Posted: 2 Jul 2016, 23:42:20 UTC
Last modified: 2 Jul 2016, 23:43:44 UTC

Well, the best solution would be to have a better mix of AR's here on Beta, like it is on main. Having a full day of only VLARs for Arecibo, really screws things up, unless you manually intervene.

Had I not lowered the <ngpus> for SoG, I would by now only get the much slower Non SoG. The system had already decided that SoG was the fastest, in fact I didn't have even one non SoG, until this Arecibo VLAR storm started.

And if I had not intervened, the APR of SoG would be so low, that I would only get non SoG, and by then the VLAR storm would be over, so the non SoG would not drop its APR. Of course leading to that the system would only give me non SoG, even on normal ARs, until the next VLAR storm, when the APR of non SoG would drop lower than the low APR SoG had from the last VLAR storm.

And then it would start over again...

No, a better mix of Arecibo AR's on Beta, thank you.
WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.
ID: 58858 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 15 Jun 16
Posts: 43
Credit: 1,836,741
RAC: 0
Australia
Message 58859 - Posted: 3 Jul 2016, 1:19:43 UTC - in response to Message 58858.  

And then it would start over again...

Yep.


No, a better mix of Arecibo AR's on Beta, thank you.

Or a way to disable the manager's best application selection when on Beta.
The more applications, and the greater the mix of work that are run on given hardware, the better.
10WUs for this application, 10 for that application, 10 for the next & so on, then repeat.
Grant
Darwin NT.
ID: 58859 · Report as offensive
Linux? You're kidding me!!
Volunteer tester
Avatar

Send message
Joined: 10 Mar 12
Posts: 1601
Credit: 11,503,026
RAC: 18,345
Sweden
Message 58861 - Posted: 3 Jul 2016, 6:24:51 UTC

Oh man!!!
Is there no end to this suffering?
Thanks for all the VLARs. Can I have something else now?

LOL
WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.
ID: 58861 · Report as offensive
BetelgeuseFive
Volunteer tester

Send message
Joined: 3 Jun 12
Posts: 64
Credit: 2,418,759
RAC: 639
Netherlands
Message 58863 - Posted: 3 Jul 2016, 10:16:25 UTC - in response to Message 58854.  

Why are the new (8.13/8.14) opencl_nvidia tasks (both SoG and sah) using so much CPU time ? I have specified -use_sleep, but I still seem to need nearly an entire CPU core to feed the GPU.

Add -v 6 to tuning line. And post few results with this option ON.


First two results (both opencl_nvidia_sah):

https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24227184
https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24227503

The second one resulted in a computation error: Disk usage limit exceeded

BOINC manager shows "free, available to BOINC: 226.47 GB".
Is there some limit on the output file size that is exceeded because using -v 6 ?

I will post SoG results as soon as they are available.

Tom
ID: 58863 · Report as offensive
Richard Haselgrove
Volunteer tester

Send message
Joined: 3 Jan 07
Posts: 1444
Credit: 3,265,669
RAC: 0
United Kingdom
Message 58864 - Posted: 3 Jul 2016, 10:38:24 UTC - in response to Message 58863.  

The second one resulted in a computation error: Disk usage limit exceeded

That will be the individual workunit limit:

<rsc_disk_bound>33554432.000000</rsc_disk_bound>

or 32 MByte.
ID: 58864 · Report as offensive
BetelgeuseFive
Volunteer tester

Send message
Joined: 3 Jun 12
Posts: 64
Credit: 2,418,759
RAC: 639
Netherlands
Message 58865 - Posted: 3 Jul 2016, 11:32:17 UTC - in response to Message 58864.  

The second one resulted in a computation error: Disk usage limit exceeded

That will be the individual workunit limit:

<rsc_disk_bound>33554432.000000</rsc_disk_bound>

or 32 MByte.


Hi Richard,

Thanks for the explanation. Can I change this limit and if so, where and how ?

Tom
ID: 58865 · Report as offensive
BetelgeuseFive
Volunteer tester

Send message
Joined: 3 Jun 12
Posts: 64
Credit: 2,418,759
RAC: 639
Netherlands
Message 58866 - Posted: 3 Jul 2016, 11:36:53 UTC - in response to Message 58854.  

Why are the new (8.13/8.14) opencl_nvidia tasks (both SoG and sah) using so much CPU time ? I have specified -use_sleep, but I still seem to need nearly an entire CPU core to feed the GPU.

Add -v 6 to tuning line. And post few results with this option ON.


And here are two SoG tasks (unfortunately both errored out with "Disk usage limit exceeded". I hope they still contain some useful information.

https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24229292
https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24230098

I'm changing back to -v 1 to prevent further errors.
If you need more examples or other options please let me know.

Thank you for looking into this.

Tom
ID: 58866 · Report as offensive
Richard Haselgrove
Volunteer tester

Send message
Joined: 3 Jan 07
Posts: 1444
Credit: 3,265,669
RAC: 0
United Kingdom
Message 58867 - Posted: 3 Jul 2016, 11:54:41 UTC - in response to Message 58865.  

Can I change this limit and if so, where and how ?

I wrote that up at the main project. Message 1796590
ID: 58867 · Report as offensive
Profile Raistmer
Volunteer tester
Avatar

Send message
Joined: 18 Aug 05
Posts: 2423
Credit: 15,878,738
RAC: 0
Russia
Message 58890 - Posted: 4 Jul 2016, 21:59:44 UTC - in response to Message 58863.  

Await new version.
News about SETI opt app releases: https://twitter.com/Raistmer
ID: 58890 · Report as offensive
Profile Raistmer
Volunteer tester
Avatar

Send message
Joined: 18 Aug 05
Posts: 2423
Credit: 15,878,738
RAC: 0
Russia
Message 58898 - Posted: 6 Jul 2016, 8:59:36 UTC

With r3482 I don't see memory leak for own iGPU. Worth to check on other iGPUs before full-scale deployment: https://cloud.mail.ru/public/J5f8/HGuG3Vp4R
News about SETI opt app releases: https://twitter.com/Raistmer
ID: 58898 · Report as offensive
SusieQ
Volunteer tester

Send message
Joined: 12 Nov 10
Posts: 1102
Credit: 29,075,117
RAC: 17,402
United Kingdom
Message 58899 - Posted: 6 Jul 2016, 9:53:16 UTC

Very happy to test if you can give me instructions on what I need to do to run this app rather than the stock one :-)
ID: 58899 · Report as offensive
Linux? You're kidding me!!
Volunteer tester
Avatar

Send message
Joined: 10 Mar 12
Posts: 1601
Credit: 11,503,026
RAC: 18,345
Sweden
Message 58901 - Posted: 6 Jul 2016, 12:24:19 UTC - in response to Message 58898.  

With r3482 I don't see memory leak for own iGPU. Worth to check on other iGPUs before full-scale deployment: https://cloud.mail.ru/public/J5f8/HGuG3Vp4R

If I'm not half dead when I get home from work, I'll run it on my iGPU on main this evening. Just throwing it in, and editing my app_info.
WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.
ID: 58901 · Report as offensive
SusieQ
Volunteer tester

Send message
Joined: 12 Nov 10
Posts: 1102
Credit: 29,075,117
RAC: 17,402
United Kingdom
Message 58902 - Posted: 6 Jul 2016, 12:56:23 UTC - in response to Message 58899.  
Last modified: 6 Jul 2016, 12:56:37 UTC

Very happy to test if you can give me instructions on what I need to do to run this app rather than the stock one :-)

I took an educated guess and renamed the MB8_win_x86_SSSE3_OpenCL_Intel.aistub file to app_info.xml and copied that and the contents of Raistmer's zip file into my Beta project directory.

When I restarted BOINC Manager it has downloaded new tasks but is now only running GPU tasks. Not sure what has happened to the tasks I already had downloaded.

Obviously my guess was, at least partially, wrong. Can anyone tell me what I need to do to get the CPU tasks running again please?

Cheers

SusieQ
ID: 58902 · Report as offensive
Zalster
Volunteer tester

Send message
Joined: 30 Dec 13
Posts: 258
Credit: 12,167,465
RAC: 0
United States
Message 58903 - Posted: 6 Jul 2016, 13:34:43 UTC

I would volunteer information if I had it, but I only run GPU work for some time now.

Sorry.

I tend to manually edit my app_info.xml

I copy what i have from main and paste it into a notepad and then change the different parts of it to match Raistmer's new versions.
ID: 58903 · Report as offensive
Previous · 1 . . . 52 · 53 · 54 · 55 · 56 · 57 · 58 . . . 99 · Next

Message boards : News : SETI@home v8 beta to begin on Tuesday


 
©2019 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.