V10/11 of modified SETI MB CUDA + opt AP package for full multi-GPU+CPU use

[Raistmer]

"team" packs (V10 included) became obsolete now. To simplify life for users who can't read whole this thread to stay tuned with latest  changes (not too latest already though, BOINC 6.6.20 exists as recommended version more than month already AFAIK) I lock this thread .
All my own support of "teamed" packs now ceased (until I get some example why "teamed" pack should be used instead of BOINC 6.6.20 native GPU scheduling at least)

"teamed" packs were temporary solution and well served their aim during lifespan. But now this period is over.

What should be used instead:
1) BOINC 6.6.20
2) latest CUDA MB build (can be found in this thread and will be posted in new thread devoted optimized CUDA MultiBeam app).
3) appropriate app_info.xml
some examples can be found in this thread: http://setiathome.berkeley.edu/forum_thread.php?id=52589

2) and 3) could be replaced by using Jason's Lunatics installer (currently in beta stage, can be found in beta area of this forum:  http://lunatics.kwsn.net/installer-testing/index.0.html )

What should not be used
1) ncpus field in cc_config.xml
2) app_info.xml supplied with obsolete "team" packs.
3)  "teamed" modification of AK_v8 (was part of "team" packs).


The key difference from V9 packs is :
The "team" mod now supports multi-GPU configs.
There will be number_of_GPUs file in SETI project directory. By default it contains number of 1.
If you have more GPUs just edit that file (enter number of GPUs installed in host instead of 1 ).
And don't forget to change your cc_config.xml
Now ncpus value required to be NUMBER_OF_LOGICAL_PROCESSORS+NUMBER_OF_GPUs for host.
For example, for Quad with 2 CUDA-enabled GPUs it should be 6
Here is example of minimal cc_config.xml file you need
<cc_config>
<options>
<ncpus>NUMBER_OF_LOGICAL_PROCESSORS+NUMBER_OF_GPUs</ncpus>
</options>
</cc_config>
File should be placed in boinc data directory (the one that contains projects subdirectory).

Currently V10 available only in SSSE3 version, other builds will follow.

When you post to this thread seeking for help, please, don't forget to provide link on your host and description of your config (OS, number of GPU cards, what pack you use, video driver version).
It will wastly decrease number of unneeded questions and save time both for you and anyone who will help you.
Please, don't forget to check ALL STUFF (including these packs) you downloaded from Internet with updated antivirus.


Fixed version posted, please, update your configs.
Thanks mr.kjellen from SETI main forums for bug report.

[Jason00]

Thanks I downloaded this yesterday! So far no problems. I only have astropulse in my waiting WU's right now. So can't say anything of the GPU crunching.

Also will this optimized client work with any of the new Bonic beta's? I'm still running 6.4.5 waiting on word if I can upgrade it yet.

[PatrickV2]

I just registered here (after freeloading on the optimized clients posted here before), and am currently running the V10 package on my Q6600/8800GTX machine, under WinVistax86_Ultimate.

Looks to run ok, 4 process on the Q6600, a fifth called 'MB_6.08_mod_CUDA...' is taking about 5% CPU, and I see one of the 5 units running progressing at about twice the rate of the others (1 is at ~31%, 4 at ~15%).

Is there a way to see from within BOINC manager which WU is running on the GPU? (Besides observing it progresses quicker?)

Regards, Patrick.

[mr.mac52]

I checked my system with V10a X64 installed and I have finally received and processed several CUDA jobs now without errors and one did detect and kill a VLAR workunit.

I'd say your V10a x86 and x64 packages are both working as desired.

Thanks Raistmer once again for your excellent work!

[gaulois952]

I only have that in my cc_config

<cc_config>
    <log_flags>
    </log_flags>
    <options>
        <dont_contact_ref_site>1</dont_contact_ref_site>
    </options>
</cc_config>
<!-- View http://boinc.berkeley.edu/trac/wiki/ClientMessages for full set of
   options and some explanations -->

Help me

[Raistmer]

I only have that in my cc_config

<cc_config>
    <log_flags>
    </log_flags>
    <options>
        <dont_contact_ref_site>1</dont_contact_ref_site>
    </options>
</cc_config>
<!-- View http://boinc.berkeley.edu/trac/wiki/ClientMessages for full set of
   options and some explanations -->

Help me

If you don't understand cc_config structure just replace your with that from first post. And don't forget to replace yellow line with acual number.

[Leopoldo]

When you post to this thread seeking for help, please, don't forget to provide link on your host and description of your config (OS, number of GPU cards, what pack you use, video driver version).
It will wastly decrease number of unneeded questions and save time both for you and anyone who will help you.

Hello, Raistmer!

My computer http://setiathome.berkeley.edu/show_host_detail.php?hostid=4632306 works with Win2003 32-bit, Core2Duo, 1 GPU 9600GT, drv 182.06, BM 6.6.11, ncpus=3

I'm doing micromanagement of task queue at Boinc Manager everyday as You supposed at http://lunatics.kwsn.net/gpu-crunching/ak-v8-cuda-mb-team-work-mod.msg13293.html#msg13293 (1 - Suspend all tasks; 2 - resume non-VLAR task; 3 - resume VLAR tasks)

For my 2-core system normal behaviour is 3 copies of AK_... (2 running + 1 idle) and 1 copy of MB_...

Previous team packs worked as described above. But situation changed with new V10a pack! Every suspending of one running task and resuming of other task leads to one more copy of remaining in memory process "AK_v8b_win_SSSE3x_GPU_CPU_team_V10.exe".

I.e. if I wants to reassign one CPU-task to VLAR, I'll do suspending all not-running tasks, then suspending one chosen CPU-task, then resuming one VLAR-task.  After this I can see in memory already 4 copies of AK_... (2 running, 2 idle) and 1 copy of MB_... Next suspendig/resuming leads to 5 copies of AK_... (2 running + 3 idle) and 1 copy of MB_...

Suspending/resuming of whole SETI@Home project isn't helps to remove useless copies, only BM restarting helps. What am I doing wrong?

P.S. V8 had no such behaviour.

[edit]Such behaviour isn't easily repeatable - i.e. not each reassign leads to remaining copies of AK_... It is sporadically. So far I can't reproduce it[/edit]

[P51 Mustang]

My computer has been throwing a lot of computation errors since I started using V10a x64.  I went from V8 to V10a to get the new AP units.  It is a Quad core running Vista 64 bit with driver 181.22 with a single GeForce 8500 GT.

Some of the errors are the VLAR kill, but most of them are not.  Here hs a link to a workunit that is throwing the computation error.

http://setiathome.berkeley.edu/result.php?resultid=1173667098

[Raistmer]

When you post to this thread seeking for help, please, don't forget to provide link on your host and des#biption of ur config (OS, number of GPU cards, what pack you use, video driver version).
It will wastly decrease number of unneeded questions and save time both for you and anyone who will help you.

Hello, Raistmer!

My computer http://setiathome.berkeley.edu/show_host_deTAil.php?hostid=4632306 works with Win2003 32-bit, Core2Duo, 1 GPU 9600GT, drv 182.06, BM 6.6.11, ncpus=3

I'm ›ing micromanagement of task queue at Boinc Manager everyday as You supposed at http://lunatics.kwsn.net/gpu-crunching/ak-v8-cuda-mb-team-work-mod.msg13293.html#msg13293 (1 - Suspend all tasks; 2 - resume non-VLAR task; 3 - resume VLAR tasks)

For my 2-core system normal behaviour is 3 copies of AK_... (2 running + 1 idle) and 1 copy of MB_...

Previous team packs worked as described above. But situation changed with new V10a pack! Every suspending of one running task and resuming of other task leads to one more copy of remaining in memory process "AK_v8b_win_SSSE3x_GPU_CPU_team_V10.exe".

I.e. if I wants to reassign one CPU-task to VLAR, I'll do suspending all not-running tasks, then suspending one chosen CPU-task, then resuming one VLAR-task.  After this I can see in memory already 4 copies of AK_... (2 running, 2 idle) and 1 copy of MB_... Next suspendig/resuming leads to 5 copies of AK_... (2 running + 3 idle) and 1 copy of MB_...

Suspending/resuming of whole SETI@Home project isn't helps to remove useless copies, only BM restarting helps. What am I doing wrong?

P.S. V8 had no such behaviour.

[edit]Such behaviour isn't easily repeatable - i.e. not each reassign leads to remaining copies of AK_... It is sporadically. So far I can't reproduce it[/edit]

Only ones that hold CUDA MB process should behave so.
CUDA MB will reject to stop now reducing probability that BOINC (or user ) making wrong decision and tries to leave GPU idle by running CPU-only apps.
So long as least single SETI MB task running (for single GPU host) SETI MB team can do its GPU scheduling w/o additional help from user.

[Raistmer]

My computer has been throwing a lot of computation errors since I started using V10a x64.  I went from V8 to V10a to get the new AP units.  It is a Quad core running Vista 64 bit with driver 181.22 with a single GeForce 8500 GT.

Some of the errors are the VLAR kill, but most of them are not.  Here is a link to a workunit that is throwing the computation error.

http://setiathome.berkeley.edu/result.php?resultid=1173667098

try to revert to old DLL versions (from older packs or stock).
Lookd like driver/CUDA runtime problem... If with same DLL/driver set stock app doesn't show such errors, report again please.

[Leopoldo]

Only ones that hold CUDA MB process should behave so.
CUDA MB will reject to stop now reducing probability that BOINC (or user ) making wrong decision and tries to leave GPU idle by running CPU-only apps.
So long as least single SETI MB task running (for single GPU host) SETI MB team can do its GPU scheduling w/o additional help from user.

Thanks! I understood possible sources of task-remnants during my VLAR-reassign process.
So, V10 already includes ideas from V8 mod like "GPU_watch_update_for_SSSE3.rar" and "No_suspend_on_BOINC_demand.rar".
BTW, Your work here is greatly appreciated!

[Raistmer]

SSE3 x86 and SSE3 x86 AMD-specific packs added.

ADDON:

SSE3 x64 Intel-specific pack added

[YukonTrooper]

How many cores set in xml for i7 with hyper-threading and one GPU?  5 or 9?

[Richard Haselgrove]

Nine

[xbill]

My computer has been throwing a lot of computation errors since I started using V10a x64.  I went from V8 to V10a to get the new AP units.  It is a Quad core running Vista 64 bit with driver 181.22 with a single GeForce 8500 GT.

Some of the errors are the VLAR kill, but most of them are not.  Here is a link to a workunit that is throwing the computation error.

http://setiathome.berkeley.edu/result.php?resultid=1173667098

try to revert to old DLL versions (from older packs or stock).
Lookd like driver/CUDA runtime problem... If with same DLL/driver set stock app doesn't show such errors, report again please.

Hi, i get the same error message when running Raistmer's_opt_pack_V10a_x86_SSSE3_MB_CPU_GPU_team on my Dell XPS 1530, Geforce 8600M GT, driver version 179.48 (newest laptop version from nvidia) under Vista Home Premium 32bit.
Example WU: http://setiathome.berkeley.edu/result.php?resultid=1174343986

Tried swapping cufft.dll, cudart.dll and libfftw3f-3-1-1a_upx.dll with included versions from V8 package, but had no luck, same error message.
But i had stock cuda running with no errors: http://setiathome.berkeley.edu/result.php?resultid=1162474169