.vlar WUs to NVIDIA GPUs (Problem Solved)


log in

Advanced search

Message boards : Number crunching : .vlar WUs to NVIDIA GPUs (Problem Solved)

1 · 2 · 3 · 4 . . . 5 · Next
Author Message
Profile [seti.international] Dirk Sadowski
Volunteer tester
Avatar
Send message
Joined: 6 Apr 07
Posts: 6970
Credit: 57,111,255
RAC: 22,470
Germany
Message 1268847 - Posted: 8 Aug 2012, 1:12:04 UTC

I noticed my BOINC got .vlar WUs for my NVIDIA GPU (after the weekly server maintenance).
Normally this shouldn't happen.

I guess something wrong happened during the weekly server maintenance.

I wrote an EMail to the admins that they are informed about this.


* Best regards! :-) * Sutaru Tsureku, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *
____________
BR



>Das Deutsche Cafe. The German Cafe.<

Profile Bill G
Avatar
Send message
Joined: 1 Jun 01
Posts: 340
Credit: 30,081,069
RAC: 71,128
United States
Message 1268863 - Posted: 8 Aug 2012, 2:21:28 UTC - in response to Message 1268847.
Last modified: 8 Aug 2012, 2:47:16 UTC

I noticed my BOINC got .vlar WUs for my NVIDIA GPU (after the weekly server maintenance).
Normally this shouldn't happen.

I guess something wrong happened during the weekly server maintenance.

I wrote an EMail to the admins that they are informed about this.


I am getting lots of .vlar's but none are going to my NVIDIA GPU, they are all for the CPU (when I am able to connect to the server)

[edit] well scratch that, my number two computer just got a schedule of a bunch of .vlar's some of which are scheduled for the NVIDIA card. As before, they just do not download....... [/edit]
____________

musicplayer
Send message
Joined: 17 May 10
Posts: 1363
Credit: 610,713
RAC: 1,718
Message 1268876 - Posted: 8 Aug 2012, 3:35:22 UTC
Last modified: 8 Aug 2012, 4:16:37 UTC

Why not give it a try? Possibly it may be working out.

Checking it out, two of the .vlar tasks I had in my task list completed successfully using CUDA and was uploaded.

There may be some interesting things found at times in these type of tasks. If it could be done in a faster way, results possibly could be obtained.

But, of course, the error rate could be high and definitely success is not always guaranteed doing it this way.

Profile Zeus Fab3r
Avatar
Send message
Joined: 17 Jan 01
Posts: 640
Credit: 79,790,070
RAC: 81,403
Serbia
Message 1268879 - Posted: 8 Aug 2012, 3:38:42 UTC

Just got 44 new gpu tasks, of which 30 are vlars.

08-Aug-12 05:30:44 SETI@home Scheduler request completed: got 44 new tasks
08-Aug-12 05:30:44 SETI@home [sched_op_debug] estimated total CPU job duration: 0 seconds
08-Aug-12 05:30:44 SETI@home [sched_op_debug] estimated total NVIDIA GPU job duration: 24075 seconds
08-Aug-12 05:30:46 SETI@home Started download of 07my10ac.24787.20521.15.10.123.vlar
08-Aug-12 05:30:46 SETI@home Started download of 07my10ac.24619.20521.14.10.124.vlar
08-Aug-12 05:30:46 SETI@home Started download of 07my10ac.24787.20521.15.10.125.vlar
.
.

____________

Who the hell is General Failure and why is he reading my harddisk?¿

Profile arkayn
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3544
Credit: 46,188,418
RAC: 30,974
United States
Message 1268882 - Posted: 8 Aug 2012, 3:59:16 UTC - in response to Message 1268876.

Why not give it a try? Possibly it may be working out.

Checking it out, two of the .vlar tasks I had in my task list completed successfully using CUDA and was uploaded.

There may be some interesting thing found at times in these type of tasks. If it could be done in a faster way, results possibly could be obtained.

But, of course, the error rate could be high and definitely success is not always guaranteed doing it this way.


It will be a good test of x41z.
____________

Profile BMaytum
Volunteer tester
Avatar
Send message
Joined: 3 Apr 99
Posts: 98
Credit: 3,374,625
RAC: 10
United States
Message 1268919 - Posted: 8 Aug 2012, 5:50:34 UTC - in response to Message 1268876.
Last modified: 8 Aug 2012, 6:08:05 UTC

Why not give it a try? Possibly it may be working out.

Checking it out, two of the .vlar tasks I had in my task list completed successfully using CUDA and was uploaded.

There may be some interesting things found at times in these type of tasks. If it could be done in a faster way, results possibly could be obtained.

But, of course, the error rate could be high and definitely success is not always guaranteed doing it this way.


I got some VLARS sent to my GPU too today (7Aug 2012) after the weekly scheduled outage. Two of them completed successfuly but took over 8,100 elapsed seconds; two of them terminated with Time Limit Exceeded at 10,161 seconds. Four more were in progress, with over 1hr:40mins elapsed and about 1h:05m estimated remaining but since Estimated remaining was continually Increasing (thus the elapsed + estimated would eventually exceed 10,161 sec ~2hr:48min), I manually aborted them. I'm running a GTX580 with v301.42 nVidia drivers, with count= 0.25 (thus GPU can process 4 concurrent workunits). Using BOINC v 7.0.28, and Lunatics optimized MB applications.

When I later got another cache-load of 11 WUs, I manually aborted the 4 GPU VLARs in that batch immediately, given my less-than stellar experince with the prior 6 VLARs on my GPU as noted just above.
____________
Sabertooth Z77, i7-3770K@4.2GHz, GTX680, W8.1Pro x64
P5N32-E SLI, C2D E8400@3Ghz, GTX580, Win7SP1Pro x64 & PCLinuxOS2014 x86

Kamu
Send message
Joined: 19 Jan 02
Posts: 56
Credit: 9,810,425
RAC: 36
Finland
Message 1268937 - Posted: 8 Aug 2012, 6:30:04 UTC

Yep, about 300 Nvidia vlar wu's here. : )

I'm wondering how we'll get credits with these. It's gonna take while to crunch those 300.

-Kimmo-

____________
Computers: obelix

Grant (SSSF)
Send message
Joined: 19 Aug 99
Posts: 5566
Credit: 51,402,905
RAC: 41,732
Australia
Message 1268941 - Posted: 8 Aug 2012, 6:52:15 UTC


Just had a look at my systems. A few VLARs on one NVidia card, several dozen on the other.
____________
Grant
Darwin NT.

LadyL
Volunteer tester
Avatar
Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1268986 - Posted: 8 Aug 2012, 10:35:30 UTC - in response to Message 1268876.
Last modified: 8 Aug 2012, 10:53:51 UTC

Why not give it a try? Possibly it may be working out.

Checking it out, two of the .vlar tasks I had in my task list completed successfully using CUDA and was uploaded.

There may be some interesting things found at times in these type of tasks. If it could be done in a faster way, results possibly could be obtained.

But, of course, the error rate could be high and definitely success is not always guaranteed doing it this way.


This one time you are getting the polite version of my reply, because other people, especially those fairly new to the project, may not know why a 'no VLAR to NVidia GPU' policy and code was established:

VLAR run incredibly slow with stock (6.08/6.09/6.10) apps. They can run so badly that the whole system freezes or outright crashes.
That was true with drivers two years ago - I don't think anybody ever established whether new drivers cope better and I'm pretty sure nobody wants to really try.

So at some point code was introduced into the scheduler to mark tasks below a certain AR as VLAR and not to send them to NVidia GPUs.

BTW, for the past few releases and RC x41z, optimised apps have not had that problem (system freeze) - but as has been shown in thread VLAR still run slower (a lot slower/too slow) - iirc at the other end of the processing spectrum from VHAR/shorties there just might be more to precess.
Since we hope v7 MB is still on the agenda and thus the GPU app will eventually become x41z, getting rid of a whole host of frist generation app problems including -12. We hadn't made up our collective mind yet if we had a case to lift the policy - you'd need production hosts testing real life performance on VLAR for a good statistical data basis.

ATM I'd rather have the restriction back - better to err on the side of caution.

NB If you have receieved VLAR and you start getting -177/-197 best to run Fred's rescheduler to extend the time limit, provided you are happy to let them process at such a slow pace.
____________
I'm not the Pope. I don't speak Ex Cathedra!

juan BFB
Volunteer tester
Avatar
Send message
Joined: 16 Mar 07
Posts: 4618
Credit: 233,530,268
RAC: 345,410
Brazil
Message 1268989 - Posted: 8 Aug 2012, 11:27:41 UTC - in response to Message 1268986.
Last modified: 8 Aug 2012, 11:40:55 UTC

Just kiddin (with a little of true): Help us LadyL you and your team are our only hope!

As i remeembr, A single Vlar takes more than an hour to process on a 580 against the 12 min of a "normal" WU...
____________

LadyL
Volunteer tester
Avatar
Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1269004 - Posted: 8 Aug 2012, 11:52:12 UTC

IIRC the no vlar is a switch in the sheduler, that might have become lost when Eric upgraded - we've had that happen before.

Richard is working on 'how to get the server to resend the VLAR to the CPU' instructions, which he will post once he's confirmed the procedure works reliably.
____________
I'm not the Pope. I don't speak Ex Cathedra!

Richard Haselgrove
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8275
Credit: 44,963,960
RAC: 13,832
United Kingdom
Message 1269021 - Posted: 8 Aug 2012, 12:22:10 UTC - in response to Message 1269004.

Richard is working on 'how to get the server to resend the VLAR to the CPU' instructions, which he will post once he's confirmed the procedure works reliably.

OK - as the Lady says...

First, these instructions are a first draft, and pretty telegraphic. They assume you're already familiar with the terminology, you know where to find the various BOINC files, and you know the rules for making changes to them. That's what we used to call ADVANCED USERS ONLY.

That's the only warning you're going to get. Read the instructions through carefully: check that you understand every point, and how to do it. If you're at all uncomfortable, don't even start. You're on your own from here.

  • Ensure you have a CPU application active for MB tasks
  • Unset 'Use NV GPU' (web preferences)
  • Set 'Use CPU' (web preferences)
  • Set 'No new tasks' (BOINC Manager)
  • Update project (BOINC Manager - if needed, some versions will report work immediately when NNT is set)
  • Suspend networking
  • Stop BOINC
  • Make backup copy of all .vlar datafiles
  • Edit client_state.xml: remove all '<result>' blocks for .vlar tasks
  • Restart BOINC
  • Restore all .vlar datafiles
  • Resume networking
  • Allow new work
  • Wait until all VLAR work has been resent to CPU
  • Set 'Use NV GPU' (web preferences)
  • Rinse and repeat

Kamu
Send message
Joined: 19 Jan 02
Posts: 56
Credit: 9,810,425
RAC: 36
Finland
Message 1269031 - Posted: 8 Aug 2012, 13:01:32 UTC

Thank You Richard!

It looks like everything went ok, resent, 287 VLARs.

-Kimmo-

____________
Computers: obelix

Profile [seti.international] Dirk Sadowski
Volunteer tester
Avatar
Send message
Joined: 6 Apr 07
Posts: 6970
Credit: 57,111,255
RAC: 22,470
Germany
Message 1269039 - Posted: 8 Aug 2012, 13:31:46 UTC - in response to Message 1268847.
Last modified: 8 Aug 2012, 13:35:59 UTC

Sutaru Tsureku wrote:
(...)
I wrote an EMail to the admins that they are informed about this.


Just for info (because I got request)...

I EMailed Dave, Eric, Matt and Jeff.

I got response from Eric, that he will inform/ask Jeff.

So they are aware of the problem and will solve it so fast as possible.


* Best regards! :-) * Sutaru Tsureku, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. *
____________
BR



>Das Deutsche Cafe. The German Cafe.<

Profile ignorance is no excuse
Avatar
Send message
Joined: 4 Oct 00
Posts: 9529
Credit: 44,432,240
RAC: 132
Korea, North
Message 1269050 - Posted: 8 Aug 2012, 13:48:03 UTC - in response to Message 1269039.

FOr now I'd suggest getting the BOINC rescheduler and have those VLAR's rescheduled to your CPU. You'll still have the WU's onboard and they won't have to be resent by S@H when they fail on your Nvidia card.
____________
In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

End terrorism by building a school

Richard Haselgrove
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8275
Credit: 44,963,960
RAC: 13,832
United Kingdom
Message 1269054 - Posted: 8 Aug 2012, 14:00:34 UTC - in response to Message 1269050.

FOr now I'd suggest getting the BOINC rescheduler and have those VLAR's rescheduled to your CPU. You'll still have the WU's onboard and they won't have to be resent by S@H when they fail on your Nvidia card.

But you are likely to mess up the server's averaging and credit-granting records. I can't be bothered to work out whether you're likely to request too much credit for yourself (only to be dragged back down by your wingmate), or to request too little and drag your wingmate down with you.

The purpose of my 'resend' recipe was to get the server records updated to show the tasks allocated to CPU - that way, runtime and credit should be accurate.

Profile Fred E.
Volunteer tester
Send message
Joined: 22 Jul 99
Posts: 731
Credit: 22,132,295
RAC: 25,308
United States
Message 1269058 - Posted: 8 Aug 2012, 14:23:00 UTC

I noticed that Eric (or someone) added a "Use ATI GPU" preference to the project preference page, probably for the new AP - for - ATI application. I don't have an ATI GPU, but it was set to on. I turned it off and haven't got a VLAR for Nvidia on the last 5 successful gpu work requests.

Can't be certain there's cause and effect here, but if you don't have an ATI card, you might as well turn it off.
____________
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.

Richard Haselgrove
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8275
Credit: 44,963,960
RAC: 13,832
United Kingdom
Message 1269060 - Posted: 8 Aug 2012, 14:28:21 UTC - in response to Message 1269058.

I noticed that Eric (or someone) added a "Use ATI GPU" preference to the project preference page ...

It's in the standard BOINC web code, but it's hidden (because presumed useless) until there's an ATI application available for stock download. It will have appeared automatically when Eric added the ATI AP app last night (see news), but it's a good point about checking that the default values on your account are right for you.

Kevin Olley
Send message
Joined: 3 Aug 99
Posts: 368
Credit: 34,439,812
RAC: 5,884
United Kingdom
Message 1269117 - Posted: 8 Aug 2012, 16:03:07 UTC - in response to Message 1268986.

This one time you are getting the polite version of my reply, because other people, especially those fairly new to the project, may not know why a 'no VLAR to NVidia GPU' policy and code was established:

VLAR run incredibly slow with stock (6.08/6.09/6.10) apps. They can run so badly that the whole system freezes or outright crashes.
That was true with drivers two years ago - I don't think anybody ever established whether new drivers cope better and I'm pretty sure nobody wants to really try.



I did, My GPU's (470's) will run VLAR's at 1 WU per card without any problems but is slow, 2 WU per card causes lag and 3 WU per card causes major problems.

I tried it when we had download problems a while ago.


____________
Kevin


Richard Haselgrove
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8275
Credit: 44,963,960
RAC: 13,832
United Kingdom
Message 1269192 - Posted: 8 Aug 2012, 18:35:12 UTC

Hmmm. My recipe worked very nicely while I only had a few VLARs, and they'd all arrived in a neat contiguous block. But now I've got a boatload more, and they're all dotted around individually in ones and twos.

Does anyone know of a nice automated way of finding/deleting a block like this?

<result>
<name>20fe12ag.23781.23954.10.10.174.vlar_2</name>
<final_cpu_time>0.000000</final_cpu_time>
<final_elapsed_time>0.000000</final_elapsed_time>
<exit_status>0</exit_status>
<state>2</state>
<platform>windows_intelx86</platform>
<version_num>610</version_num>
<plan_class>cuda_fermi</plan_class>
<wu_name>20fe12ag.23781.23954.10.10.174.vlar</wu_name>
<report_deadline>1348464226.000000</report_deadline>
<received_time>1344446227.080149</received_time>
<file_ref>
<file_name>20fe12ag.23781.23954.10.10.174.vlar_2_0</file_name>
<open_name>result.sah</open_name>
</file_ref>
</result>

- in other words, everything from <result> to </result> inclusive, but ONLY when they contain both vlar and version 610 - I don't want to delete the ones I've already had resent...

1 · 2 · 3 · 4 . . . 5 · Next

Message boards : Number crunching : .vlar WUs to NVIDIA GPUs (Problem Solved)

Copyright © 2014 University of California