WARNING: This website is obsolete! Please follow this link to get to the new Albert@Home website!

[New release] BRP app v1.22 feedback thread

Message boards : Problems and Bug Reports : [New release] BRP app v1.22 feedback thread
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Profile Oliver Behnke
Volunteer moderator
Project administrator
Project developer

Send message
Joined: 4 Sep 07
Posts: 130
Credit: 8,545,955
RAC: 0
Message 111876 - Posted: 27 Feb 2012, 15:45:46 UTC
Last modified: 2 Mar 2012, 9:20:57 UTC

Hi,

We just released BRP4 v1.22 which adds a number of improvements.

Notes:
* OpenCL GPU memory requirements reduced by 128 MB (to ~360 MB)
* Dropped support for OpenCL 1.0 GPUs (in favor of the above, OpenCL 1.0 only GPUs like the Radeon 4xxx are too slow anyway)
* More graceful OpenCL memory error handling
* Known issue: no OpenCL support for Mac OS X for the time being (we're looking into a potential Apple bug, no pun intended)
* CUDA apps are enabled again for the time being
* Input data download volume (per work unit) can be reduced by a factor of two (soon)
* Please use the latest Catalyst driver (>=12.1) and BOINC client (>=7.0.12).

Let's try and collect your feedback to this specific release (and this one only) in this thread.


Thanks,
Oliver
ID: 111876 · Report as offensive     Reply Quote
Profile mickydl*

Send message
Joined: 8 Dec 11
Posts: 6
Credit: 6,000
RAC: 0
Message 111878 - Posted: 27 Feb 2012, 20:12:35 UTC - in response to Message 111876.  
Last modified: 27 Feb 2012, 20:14:50 UTC

Just to be sure: By "latest Catalyst driver (>=12.1)" you mean "AMD Catalystâ„¢ 11.11 - Revision number 12.1" ?

I can't find any Catalyst 12.1 for Linux on the AMD site.

mickydl*
ID: 111878 · Report as offensive     Reply Quote
Profile Trog Dog
Avatar

Send message
Joined: 25 Nov 05
Posts: 204
Credit: 64,008
RAC: 0
Message 111879 - Posted: 28 Feb 2012, 0:29:12 UTC - in response to Message 111878.  

ID: 111879 · Report as offensive     Reply Quote
skildude

Send message
Joined: 15 Nov 11
Posts: 9
Credit: 103,497
RAC: 0
Message 111880 - Posted: 28 Feb 2012, 2:37:36 UTC

should we dump Wu's still using the 1.21 app? I've noticed a few inconclusives and errors on that app. not seeing any problems from other GPU projects
ID: 111880 · Report as offensive     Reply Quote
Profile Oliver Behnke
Volunteer moderator
Project administrator
Project developer

Send message
Joined: 4 Sep 07
Posts: 130
Credit: 8,545,955
RAC: 0
Message 111881 - Posted: 28 Feb 2012, 9:03:25 UTC - in response to Message 111878.  

Just to be sure: By "latest Catalyst driver (>=12.1)" you mean "AMD Catalystâ„¢ 11.11 - Revision number 12.1" ?

I can't find any Catalyst 12.1 for Linux on the AMD site.


That's just because of AMD's sloppy web editing. The one you found is the 12.1 driver.

Oliver
ID: 111881 · Report as offensive     Reply Quote
Profile Oliver Behnke
Volunteer moderator
Project administrator
Project developer

Send message
Joined: 4 Sep 07
Posts: 130
Credit: 8,545,955
RAC: 0
Message 111882 - Posted: 28 Feb 2012, 9:05:38 UTC - in response to Message 111880.  
Last modified: 28 Feb 2012, 9:06:03 UTC

should we dump Wu's still using the 1.21 app? I've noticed a few inconclusives and errors on that app. not seeing any problems from other GPU projects


Depends on what errors you saw. If they are memory-related you probably want to reset your project. You'll be resent the same tasks but they'll be crunched with the latest app version which required less memory (OpenCL only).

Cheers,
Oliver
ID: 111882 · Report as offensive     Reply Quote
Profile pragmatic prancing periodic problem child, left
Avatar

Send message
Joined: 26 Jan 05
Posts: 1639
Credit: 70,000
RAC: 0
Message 111883 - Posted: 28 Feb 2012, 23:47:36 UTC

Task comes in with <rsc_fpops_est>300000000000000.000000</rsc_fpops_est> which tells BOINC the task is going to take 210 hours and a bit, so BOINC will run it for a long time in panic mode. Can we please get a reasonable fpops estimate, one that doesn't immediately throw Albert tasks in High Priority?
Jord.

BOINC FAQ Service

They say most of your brain shuts down in cryo-sleep. All but the primitive side, the animal side. No wonder I'm still awake.
ID: 111883 · Report as offensive     Reply Quote
spingadus[MM]

Send message
Joined: 15 Oct 06
Posts: 4
Credit: 250,000
RAC: 0
Message 111884 - Posted: 29 Feb 2012, 2:18:48 UTC
Last modified: 29 Feb 2012, 2:19:26 UTC

Are there actually tasks available? I'm not receiving anything for my GPU.

HD 6970
BOINC 7.0.18
ATI Driver 12.1


16459 Albert@Home 2/28/2012 6:15:47 PM update requested by user
16460 Albert@Home 2/28/2012 6:15:48 PM Sending scheduler request: Requested by user.
16461 Albert@Home 2/28/2012 6:15:48 PM Requesting new tasks for ATI
16462 Albert@Home 2/28/2012 6:15:51 PM Scheduler request completed: got 0 new tasks
16463 Albert@Home 2/28/2012 6:15:51 PM No tasks sent

I've even suspended all tasks before trying to update.
ID: 111884 · Report as offensive     Reply Quote
skildude

Send message
Joined: 15 Nov 11
Posts: 9
Credit: 103,497
RAC: 0
Message 111885 - Posted: 29 Feb 2012, 2:22:12 UTC - in response to Message 111884.  

you have to leave a cpu free otherwise it's unlikely to get work.
ID: 111885 · Report as offensive     Reply Quote
spingadus[MM]

Send message
Joined: 15 Oct 06
Posts: 4
Credit: 250,000
RAC: 0
Message 111887 - Posted: 29 Feb 2012, 3:06:38 UTC - in response to Message 111885.  

Freed up 2 threads out of 8 and still nothing. Still no work. I'm currently running Moo! so I suspended the project as well. No tasks.
ID: 111887 · Report as offensive     Reply Quote
Profile Oliver Behnke
Volunteer moderator
Project administrator
Project developer

Send message
Joined: 4 Sep 07
Posts: 130
Credit: 8,545,955
RAC: 0
Message 111888 - Posted: 29 Feb 2012, 9:48:09 UTC - in response to Message 111884.  
Last modified: 29 Feb 2012, 9:54:46 UTC

Are there actually tasks available? I'm not receiving anything for my GPU.


Yes there are and your config looks fine so far. Please have a look at the BOINC event log: did BOINC recognize your AMD GPU as OpenCL device? According to our logs it doesn't seem to be the case. You might need to reinstall the Catalyst driver. Also, remember to start the X server and make sure that BOINC can access the X display. If "clinfo" exists on your system you may use it to verify that your GPU is properly enumerated by OpenCL.

Cheers,
Oliver
ID: 111888 · Report as offensive     Reply Quote
spingadus[MM]

Send message
Joined: 15 Oct 06
Posts: 4
Credit: 250,000
RAC: 0
Message 111889 - Posted: 29 Feb 2012, 10:15:52 UTC - in response to Message 111888.  

No X windows here, I'm running win7.

Is there somewhere in the logs that will show me if the card is enumerated correctly?

I re-installed the 12.1 driver yesterday and did a custom install instead of express. Everything appeared to be checked.

Its 2am here and I'm sleep typing, so I'll check when I wake up. :P
ID: 111889 · Report as offensive     Reply Quote
Profile Oliver Behnke
Volunteer moderator
Project administrator
Project developer

Send message
Joined: 4 Sep 07
Posts: 130
Credit: 8,545,955
RAC: 0
Message 111890 - Posted: 29 Feb 2012, 10:50:17 UTC - in response to Message 111889.  

No X windows here, I'm running win7


Oops, sorry :-) When you start the BOINC client it'll list all GPUs in the event log (advanced view). For AMD/ATI devices it might talk about CAL and OpenCL - we're interested only in the latter. You should find the list of GPUs more or less at the top of the event log, right before the registered projects are mentioned.

Oliver
ID: 111890 · Report as offensive     Reply Quote
Profile mickydl*

Send message
Joined: 8 Dec 11
Posts: 6
Credit: 6,000
RAC: 0
Message 111891 - Posted: 29 Feb 2012, 14:25:12 UTC

My first OpenCL WU with the v1.22 app validated aigainst a CUDA result. No problem (see this Task). However, I believe that I am using an earlier version of Catalyst (< 12.1). I'll have to check that when I'm home from work.

mickydl*
ID: 111891 · Report as offensive     Reply Quote
oz

Send message
Joined: 28 Feb 05
Posts: 10
Credit: 1,285,478
RAC: 0
Message 111892 - Posted: 1 Mar 2012, 8:01:51 UTC

This happens in 7.0.18:
The scheduler requests new jobs, and 3 seconds later it starts S6LV1. I can not imagine that the download of the task is completed then? Normally we can see something like this
Started download of p2030.20111110.G39.19-00.79.N.b2s0g0.00000_3648.binary
Finished download of p2030.20111110.G39.19-00.79.N.b2s0g0.00000_3648.binary

first.

01-Mar-2012 08:39:39 [Albert@Home] Sending scheduler request: To fetch work.
01-Mar-2012 08:39:39 [Albert@Home] Reporting 4 completed tasks, requesting new tasks for CPU
01-Mar-2012 08:39:48 [Albert@Home] Scheduler request completed: got 1 new tasks
01-Mar-2012 08:39:48 [Albert@Home] Resent lost task h1_0059.95_S6GC1__39_S6LV1A_1
01-Mar-2012 08:39:51 [Albert@Home] Starting task h1_0059.95_S6GC1__39_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 10
01-Mar-2012 08:39:52 [Albert@Home] Computation for task h1_0059.95_S6GC1__39_S6LV1A_1 finished
01-Mar-2012 08:39:52 [Albert@Home] Output file h1_0059.95_S6GC1__39_S6LV1A_1_0 for task h1_0059.95_S6GC1__39_S6LV1A_1 absent
01-Mar-2012 08:41:39 [Albert@Home] Sending scheduler request: To fetch work.
01-Mar-2012 08:41:39 [Albert@Home] Reporting 1 completed tasks, requesting new tasks for CPU
01-Mar-2012 08:41:41 [Albert@Home] Scheduler request completed: got 4 new tasks
01-Mar-2012 08:41:43 [Albert@Home] Starting task h1_0059.95_S6GC1__35_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 10
01-Mar-2012 08:41:43 [Albert@Home] Starting task h1_0059.95_S6GC1__33_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 11
01-Mar-2012 08:41:43 [Albert@Home] Starting task h1_0059.95_S6GC1__34_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 12
01-Mar-2012 08:41:44 [Albert@Home] Computation for task h1_0059.95_S6GC1__35_S6LV1A_1 finished
01-Mar-2012 08:41:44 [Albert@Home] Output file h1_0059.95_S6GC1__35_S6LV1A_1_0 for task h1_0059.95_S6GC1__35_S6LV1A_1 absent
01-Mar-2012 08:41:44 [Albert@Home] Starting task h1_0059.95_S6GC1__36_S6LV1A_1 using einstein_S6LV1 version 110 (SSE2) in slot 10
01-Mar-2012 08:41:45 [Albert@Home] Computation for task h1_0059.95_S6GC1__33_S6LV1A_1 finished
01-Mar-2012 08:41:45 [Albert@Home] Output file h1_0059.95_S6GC1__33_S6LV1A_1_0 for task h1_0059.95_S6GC1__33_S6LV1A_1 absent
01-Mar-2012 08:41:46 [Albert@Home] Computation for task h1_0059.95_S6GC1__34_S6LV1A_1 finished
01-Mar-2012 08:41:46 [Albert@Home] Output file h1_0059.95_S6GC1__34_S6LV1A_1_0 for task h1_0059.95_S6GC1__34_S6LV1A_1 absent
01-Mar-2012 08:41:47 [Albert@Home] Computation for task h1_0059.95_S6GC1__36_S6LV1A_1 finished
01-Mar-2012 08:41:47 [Albert@Home] Output file h1_0059.95_S6GC1__36_S6LV1A_1_0 for task h1_0059.95_S6GC1__36_S6LV1A_1 absent


ID: 111892 · Report as offensive     Reply Quote
Profile Oliver Behnke
Volunteer moderator
Project administrator
Project developer

Send message
Joined: 4 Sep 07
Posts: 130
Credit: 8,545,955
RAC: 0
Message 111893 - Posted: 1 Mar 2012, 9:40:42 UTC - in response to Message 111892.  


The scheduler requests new jobs, and 3 seconds later it starts S6LV1.


Well, S6LV1 tasks are not the same as BRP tasks. S6LV1 tasks re-use data already present on your host while BRP data is only used once, for a single WU. Looking at the error output of the failed S6LV1 task should tell us what happened.

Please open another thread for that problem if it persists. This thread is meant to discuss BRP v1.22 only.

Cheers,
Oliver
ID: 111893 · Report as offensive     Reply Quote
skildude

Send message
Joined: 15 Nov 11
Posts: 9
Credit: 103,497
RAC: 0
Message 111894 - Posted: 2 Mar 2012, 5:13:13 UTC - in response to Message 111890.  
Last modified: 2 Mar 2012, 5:14:45 UTC

No X windows here, I'm running win7


Oops, sorry :-) When you start the BOINC client it'll list all GPUs in the event log (advanced view). For AMD/ATI devices it might talk about CAL and OpenCL - we're interested only in the latter. You should find the list of GPUs more or less at the top of the event log, right before the registered projects are mentioned.

Oliver

He may need to uninstall the drivers, run driver sweep, then reinstall the 12.1 drivers. Leaving old drivers wreaks havoc on the OpenCL apps at Seti. Probably the same here. His Card is recognized as a 6900 series and he is running the 7.0.18 BOINC so that isn't a problem.

Could running vbox be a problem?
ID: 111894 · Report as offensive     Reply Quote
Profile Oliver Behnke
Volunteer moderator
Project administrator
Project developer

Send message
Joined: 4 Sep 07
Posts: 130
Credit: 8,545,955
RAC: 0
Message 111895 - Posted: 2 Mar 2012, 9:22:51 UTC - in response to Message 111894.  
Last modified: 2 Mar 2012, 9:23:37 UTC


Could running vbox be a problem?


Don't know but could be if vbox acquires the GPU somehow...

Oliver
ID: 111895 · Report as offensive     Reply Quote
Profile pragmatic prancing periodic problem child, left
Avatar

Send message
Joined: 26 Jan 05
Posts: 1639
Credit: 70,000
RAC: 0
Message 111896 - Posted: 2 Mar 2012, 15:15:01 UTC - in response to Message 111895.  

No it doesn't. No GPUs are being used on T4T or the Vboxwrapper test project (the only two projects at this time where VBox is being used), other than for showing graphics of sorts. And then these projects require Vbox 4.1.4 or higher, as far as I know.

I'll go with driver corruption as well. It certainly never hurts to completely clean out previous drivers and then reinstall any later as new.
Jord.

BOINC FAQ Service

They say most of your brain shuts down in cryo-sleep. All but the primitive side, the animal side. No wonder I'm still awake.
ID: 111896 · Report as offensive     Reply Quote
Profile pragmatic prancing periodic problem child, left
Avatar

Send message
Joined: 26 Jan 05
Posts: 1639
Credit: 70,000
RAC: 0
Message 111897 - Posted: 3 Mar 2012, 1:56:53 UTC - in response to Message 111883.  

Tasks still come in expecting to run for 205 hours.
So I still have Albert tasks running in a panic.

03/03/2012 02:47:11 | Albert@Home | [rr_sim] Result p2030.20111110.G39.19-00.79.N.b3s0g0.00100_864_3 projected to miss deadline.
03/03/2012 02:47:11 | Albert@Home | [rr_sim] Project has 1 projected ATI deadline misses
03/03/2012 02:47:31 | Albert@Home | [rr_sim] p2030.20111110.G39.19-00.79.N.b3s0g0.00100_864_3 misses deadline by 614511.77

<time_stats>
    <on_frac>0.939516</on_frac>
    <connected_frac>0.783900</connected_frac>
    <active_frac>0.392607</active_frac>
    <gpu_active_frac>0.392447</gpu_active_frac>
    <last_update>1330725382.604116</last_update>
</time_stats>


Of course, it's because BOINC thinks that the 205 hours it's estimated to go do is really 205h / (39 / 100) = 525h (or almost 22 days). A tad difficult to do in 14 days. So it'll run from start to finish in high priority. And as we can see in here, DCF is no longer really used with Boinc 7. Not that it matters, DCF is 7.5, way too high to use reliably.

So, pretty please, can the fpops estimate be adjusted enough that they don't come in thinking to take 200+ hours?
Jord.

BOINC FAQ Service

They say most of your brain shuts down in cryo-sleep. All but the primitive side, the animal side. No wonder I'm still awake.
ID: 111897 · Report as offensive     Reply Quote
1 · 2 · 3 · Next

Message boards : Problems and Bug Reports : [New release] BRP app v1.22 feedback thread



This material is based upon work supported by the National Science Foundation (NSF) under Grant PHY-0555655 and by the Max Planck Gesellschaft (MPG). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the investigators and do not necessarily reflect the views of the NSF or the MPG.

Copyright © 2019 Bruce Allen for the LIGO Scientific Collaboration