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

[n7rfa]

Your app_info is from the team mod, it doesn't work with 6.6.*

V10 pack does work with any 6.6.x BOINC including 6.6.20. Actually I running it exactly with BOINC 6.6.20 right now.
But starting from BOINC 6.6.14 as was mentioned many times in this thread already there is possibility to use BOINC own scheduling mechanism for CUDA applications to run both CPU and GPU SETI MB versions simultaneously. That's why team pack can be disabled now.
But to do this you need replace app_info.xml _and_ SETI CPU MB "teamed" app ("teamed" AK_v8). Only SETI CUDA MB and SETI AP should be leaved untouched while upgrading from V10 pack to BOINC 6.6.20.
You need to delete cc_config.xml also (or just edit it to remove ncpus section if you use cc_config for some other purposes too).

What with 23 pages of app_info.xml postings, corrections, BOINC version x.y.z and a.b.c, it is more then just a bit confusing.

I wasn't running CUDA before. I just replaced a GPU that was failing and upgraded to a CUDA GPU. So there were no changes to be made to the cc_config.xml file.

In anticipation of the CUDA work, I upgraded from a 5.x BOINC to the latest version 6.6.20.

I used the ap_info.xml from the v10a package and basically only removed the AP information. After that, I just copied everything into the projects folder and started BOINC back up.

Exactly what other changes do I need to make to the app_info.xml?

If need be, I am willing to downgrade BOINC to 6.4.7.

[Raistmer]

Downgrade not needed.
You can use first posted app_info from corresponding SETI forum thread (and more or less safely ignore all later 23 pages ) but check if app names corresponding names of your executables.
That way you need unmodded AK_v8 (from download section of this site), my V10/11 CUDA MB build (from this thread) (I assume you don't want to run AP so not listed it).

Another way is install V10 pack as it is. Add cc_config file with ncpus field (look first page). For doing ONLY SETI MB on both CPU and GPU, w/o AP or other projects V10 pack seems to me as more preferable way (still) cause it can balance workload between CPU and GPU while BOINC has separate work fetch (you can have plenty of "CPU" MB tasks while your GPU will sit idle cause it can't fetch new work fot itself).

If you participate in some another project too then BOINC 6.6.20 preferable.

[Slagathor]

I have tried a few time to get this to work......

Each time I error out and loose some WU's.......

Could I get some help please?

This is the machine:
http://setiathome.berkeley.edu/show_host_detail.php?hostid=4293313

It's running Vista Ultimate 32bit, it has 2 8800GTS's 320mb, 182.50,  Raistmer's_opt_pack_V10a_x86_SSE3_AMD_MB_CPU_GPU_team.rar, BOINC 6.6.20


Also, I am trying to just get Cuda to run on there, what I am looking for is a way to kill off VLAR WU's....

Thanks for your time...

[Raistmer]

If you lose tasks it means you don't follow the rules for upgrading.

"
The recommended way to do any (not only this one) update of opt apps:

1) DISABLE network access in BOINC
2) STOP BOINC
3) Copy BOINC whole data directory in another place as backup.
4) install update
5) start BOINC.
6) if all work fine - resume network access, else - stop BOINC and restore BOINC data folder from backup.

In other case you risk to trash whole your cache and be blamed by all few hundreds of your wingmans for that unfriendly action!
"

<core_client_version>6.6.20</core_client_version>
<![CDATA[
<message>
app_version download error: couldn't get input files:
<file_xfer_error>
  <file_name>setiathome_6.08_windows_intelx86__cuda.exe</file_name>
  <error_code>-200</error_code>
</file_xfer_error>


That means you don't have this file but you need it before downloading any work with app_info file installed.
Either edit app_info to point on CUDA MB app you really try to install or download this stock version first.

[Slagathor]

I did the data folder back up, but even when I put it back it was gone, but anywase.

I got rid of the stock app, prolly shouldn't have, and I just copied the whole rar to the work folder.

The error you posted happined a bit ago, before I said ta hell with it and used the stock app again.


But, like before, like I said, I copied the contents of the rar to the work folder, I didnt change anything in the xml, and everything errors, even with an empty cache, then it downloads a couple WU's and them error out also....

[Raistmer]

I did the data folder back up, but even when I put it back it was gone, but anywase.

I got rid of the stock app, prolly shouldn't have, and I just copied the whole rar to the work folder.

The error you posted happined a bit ago, before I said ta hell with it and used the stock app again.


But, like before, like I said, I copied the contents of the rar to the work folder, I didnt change anything in the xml, and everything errors, even with an empty cache, then it downloads a couple WU's and them error out also....

1)List content of project folder
2) Give link on fresh errored result.

[Slagathor]

I did the data folder back up, but even when I put it back it was gone, but anywase.

I got rid of the stock app, prolly shouldn't have, and I just copied the whole rar to the work folder.

The error you posted happined a bit ago, before I said ta hell with it and used the stock app again.


But, like before, like I said, I copied the contents of the rar to the work folder, I didnt change anything in the xml, and everything errors, even with an empty cache, then it downloads a couple WU's and them error out also....

1)List content of project folder
2) Give link on fresh errored result.

1) right now it's all stock.
2) I am pretty sure this is the last errored one I got with the Raistmer's_opt_pack_V10a: http://setiathome.berkeley.edu/result.php?resultid=1203461771
Thats why I am a bit confused....

I am sorry if I am being a pain, thanks again for the help

[Raistmer]

[1) right now it's all stock.
2) I am pretty sure this is the last errored one I got with the Raistmer's_opt_pack_V10a: http://setiathome.berkeley.edu/result.php?resultid=1203461771
Thats why I am a bit confused....

I am sorry if I am being a pain, thanks again for the help

the same error
<core_client_version>6.6.20</core_client_version>
<![CDATA[
<message>
app_version download error: couldn't get input files:
<file_xfer_error>
  <file_name>setiathome_6.08_windows_intelx86__cuda.exe</file_name>
  <error_code>-200</error_code>
</file_xfer_error>

</message>
]]>

Possibly you tried to use V10 CUDA MB, but you had stock app filename in app_info.
That's why they all errored.

[elec999]

On my LL I reinstalled the os. And Install V10 again. But keep on getting compusion error. And have to wait 20hours each time for more work.
Thanks

[Raistmer]

On my LL I reinstalled the os. And Install V10 again. But keep on getting compusion error. And have to wait 20hours each time for more work.
Thanks

It seems it's better to upgrade to BOINC 6.6.20 for you and use provided app_info (on SETI main) for it.
You have bad luck with this pack definitely and by refusing to follow update rules you trash tasks instead of process them... not very good tactic. Why you ignore backup stage in your experiments?

[elec999]

Where can I find the app_info on seti.
Thanks

[galadriel]

http://setiathome.berkeley.edu/forum_thread.php?id=52589

[Raistmer]

Fresh V11 SETI CUDA build for standalone/team use with single GPU-multi core systems (try to restrict boinc.exe from using first core to achieve better GPU performance )
I found no speed difference between stock CUDA runtime (2.0 probably) and 2.1 CUDA runtime I used for building this app so use that version wich gives better stability on your host (no GPU freezing).

ADDON: added build w/o affinity lock. Maybe it best suited for high performance multi-GPU systems while build with affinity lock on first core could achieve better performance on single-GPU systems (with boinc.exe excluded from first core).

ADDON2: Build with VLAR kill mod disable added.

[attachment deleted by admin]

[elec999]

I plan to install three more video cards today. a 9800GTX+, two 8800GTS. Do you think this is a good idea. I will install to my B5. It can handle four video cards. And when my XFX board returns from  RMA, I will install another 8800GT.
Thanks

[codeman05]

Hey guys, with much help from Claggy, I have the optimizations working correctly on my rig.

However, after comparing my CUDA results with a couple other guys on the SETI forum, it seems like my performance may be a little off.
Rig Specs:
i7 920 @ 4.0GHz w/ Vista x64
3x GTX285 GPU's
BOINC 6.6.20

My average CUDA WU completion time is anywhere from 7 minutes to 15 minutes...usually around 15 minutes lately.
I am crunching AP at the same time and BOINC is running 8 AP's concurrently...for a total of 11 processes.

I've been reading reports of guys doing 3-5 minute CUDA wu's with virtually the exact same setup (both proc and gpu's).  Do I just have some really "bad angle" WU's...or is this a tradeoff of running both CUDA and AP...or is this unusual?

The GPU's don't really seem to be fully utilized. I know this is a crappy way of measuring, but my GPU temps seem to be about 10-15 degree's cooler when crunching than before...again a bad example, but seems to indicate less load than before the optimizers.


Thanks for the assistance, and all of your hard work on these appss.