Refugee from the old Lunatics Site

[ausymark]

Hi All

Glad to see this site up and running. I have been trying to get my new nVidia 580 card crunching Seti work units for a good 5 months now with now luck - until finding this site.

Previously I was using the Lunatics site - but its forum is closed, and the only thing I could work out was that its older CUDA 2 apps wouldn't work with the new FERMI / nVidia 5xx cards. So I started looking for another compile by someone - I was close to giving up and then I found this site - installed the software, made a couple tweaks to the system - and now I am crunching a couple work units at a time (I know i could easily crunch 4 but this is also my personal workstation - so I had to strike a balance   )

So I can only say thank you for having this here - now I'm of to find ET at >7,000 work units a day - WOOOOT!!!

Cheers   

[arkayn]

The forum is closed, but the software is still available to the general public.

I am just rehosting the software that they produce and I alpha test for them.

[ausymark]

Interesting, as they arent showing the latest CUDA seti apps for linux, its all the 2 year old (CUDA 2* ) stuff.

Keep up the great work

[arkayn]

The Fermi CUDA app was developed by Aaron Haviland from the SETI stock app, once they get the v7 stuff worked out the Lunatics stuff will be ported to Linux as well.

I saw you name in a thread as well.

Try this build, it is an beta app built from the x34 code that Jason was working on.

[ausymark]

Thanks, got it installed, and removed all references in App_info regarding the CUDA libraries - that seemed to make it work :-)

I will let you know if i hit any issues

Cheers 

Addition : Have just noticed that its running my 580 about 4c hotter - but thats still just 55c in total.

Second Addition: Seems to be running fine, no errors. Does this have debug code in it - as it looks like the graphics card memory usage is around 280mb per task - if it has debug code in it then that footprint in the release code would be less.

[ausymark]

On the bleeding edge .....

I just upgraded to Mandriva 2011, which is running CUDA 4.0.17.

Boinc/Seti - the beta one in the message above - says:

Mon 29 Aug 2011 21:08:26 CST      Starting BOINC client version 6.10.58 for x86_64-pc-linux-gnu
Mon 29 Aug 2011 21:08:26 CST      log flags: file_xfer, sched_ops, task
Mon 29 Aug 2011 21:08:26 CST      Libraries: libcurl/7.18.0 OpenSSL/0.9.8g zlib/1.2.5 c-ares/1.5.1
Mon 29 Aug 2011 21:08:26 CST      Data directory: /home/mark/Apps/BOINC
Mon 29 Aug 2011 21:08:26 CST      Processor: 8 GenuineIntel Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz [Family 6 Model 42 Stepping 7]
Mon 29 Aug 2011 21:08:26 CST      Processor: 8.00 MB cache
Mon 29 Aug 2011 21:08:26 CST      Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf pni pclmulqd
Mon 29 Aug 2011 21:08:26 CST      OS: Linux: 2.6.38.7-desktop-1mnb2
Mon 29 Aug 2011 21:08:26 CST      Memory: 11.74 GB physical, 17.32 GB virtual
Mon 29 Aug 2011 21:08:26 CST      Disk: 1.87 TB total, 1.79 TB free
Mon 29 Aug 2011 21:08:26 CST      Local time is UTC +9 hours
Mon 29 Aug 2011 21:08:26 CST      No usable GPUs found

So either CUDA is failing, or perhaps Boinc/Seti isnt seeing it.

Note: I have tried to do the following in the App_Info file:  Not specified the libcufft/libcudart/ - 3, 4 and 4.0.17 files - all with similar results.

Any Ideas?

Cheers

[arkayn]

From that, BOINC is not seeing the CUDA card at all.

One thing about CUDA 4.0 is that is suffers from down-clocking issues, due to improper thread exits

Try making certain that the system sees the card and make certain that BOINC is in the video group.

[ausymark]

From that, BOINC is not seeing the CUDA card at all.

One thing about CUDA 4.0 is that is suffers from down-clocking issues, due to improper thread exits

Try making certain that the system sees the card and make certain that BOINC is in the video group.

OK, i removed the X11 display settings/package - CUDA was looking at the Intel Graphics chip in the i7 2600. So now Boinc is seeing the nVidia GPU.

The last problem however seems to be, that seti cant process the GPU work units - even though it knows they are there. I have tried both the current and beta Seti Cuda apps with the same result. Here is a copy of the startup:

Tue 30 Aug 2011 12:04:31 CST      Starting BOINC client version 6.10.58 for x86_64-pc-linux-gnu
Tue 30 Aug 2011 12:04:31 CST      log flags: file_xfer, sched_ops, task
Tue 30 Aug 2011 12:04:31 CST      Libraries: libcurl/7.18.0 OpenSSL/0.9.8g zlib/1.2.5 c-ares/1.5.1
Tue 30 Aug 2011 12:04:31 CST      Data directory: /home/mark/Apps/BOINC
Tue 30 Aug 2011 12:04:31 CST      Processor: 8 GenuineIntel Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz [Family 6 Model 42 Stepping 7]
Tue 30 Aug 2011 12:04:31 CST      Processor: 8.00 MB cache
Tue 30 Aug 2011 12:04:31 CST      Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf pni pclmulqd
Tue 30 Aug 2011 12:04:31 CST      OS: Linux: 2.6.38.7-desktop-1mnb2
Tue 30 Aug 2011 12:04:31 CST      Memory: 11.74 GB physical, 17.32 GB virtual
Tue 30 Aug 2011 12:04:31 CST      Disk: 1.87 TB total, 1.79 TB free
Tue 30 Aug 2011 12:04:31 CST      Local time is UTC +9 hours
Tue 30 Aug 2011 12:04:31 CST      NVIDIA GPU 0: GeForce GTX 580 (driver version unknown, CUDA version 4000, compute capability 2.0, 1536MB, 1628 GFLOPS peak)
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Found app_info.xml; using anonymous platform
Tue 30 Aug 2011 12:04:31 CST   SETI@home   URL http://setiathome.berkeley.edu/; Computer ID 5458341; resource share 100
Tue 30 Aug 2011 12:04:31 CST   SETI@home   General prefs: from SETI@home (last modified 01-May-2011 08:16:11)
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Computer location: home
Tue 30 Aug 2011 12:04:31 CST   SETI@home   General prefs: no separate prefs for home; using your defaults
Tue 30 Aug 2011 12:04:31 CST      Reading preferences override file
Tue 30 Aug 2011 12:04:31 CST      Preferences:
Tue 30 Aug 2011 12:04:31 CST         max memory usage when active: 9014.33MB
Tue 30 Aug 2011 12:04:31 CST         max memory usage when idle: 10817.20MB
Tue 30 Aug 2011 12:04:31 CST         max disk usage: 3.00GB
Tue 30 Aug 2011 12:04:31 CST         (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
Tue 30 Aug 2011 12:04:31 CST      Not using a proxy
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_09mr11af_B5_P0_00280_20110826_14280.wu_1 using astropulse_v505 version 506
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_08mr11ag_B6_P1_00375_20110801_04575.wu_3 using astropulse_v505 version 506
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_10mr11aa_B4_P0_00144_20110826_06105.wu_0 using astropulse_v505 version 506
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_10mr11aa_B4_P0_00143_20110826_06105.wu_0 using astropulse_v505 version 506
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_10mr11aa_B4_P0_00142_20110826_06105.wu_1 using astropulse_v505 version 506
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_10mr11aa_B4_P0_00141_20110826_06105.wu_1 using astropulse_v505 version 506
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_09mr11af_B5_P1_00073_20110826_03583.wu_0 using astropulse_v505 version 506
Tue 30 Aug 2011 12:04:31 CST   SETI@home   Restarting task ap_09mr11af_B5_P1_00072_20110826_03583.wu_0 using astropulse_v505 version 506

my current app_info file is as follows:

<app_info>
<app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>AK_V8_linux64_ssse3</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>603</version_num>
<file_ref>
<file_name>AK_V8_linux64_ssse3</file_name>
<main_program/>
</file_ref>
</app_version>
    <app>
        <name>astropulse_v505</name>
    </app>
    <file_info>
        <name>ap_5.06r411_sse3_linux64</name>
        <executable/>
    </file_info>
    <app_version>
        <app_name>astropulse_v505</app_name>
        <version_num>506</version_num>
        <file_ref>
            <file_name>ap_5.06r411_sse3_linux64</file_name>
            <main_program/>
        </file_ref>
    </app_version>
    <app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>setiathome-r195.x86_64-pc-linux-gnu__cuda_fermi</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>611</version_num>
<plan_class>cuda_fermi</plan_class>
<avg_ncpus>0.050000</avg_ncpus>
<max_ncpus>0.050000</max_ncpus>
<coproc>
<type>CUDA</type>
<count>0.5</count>
</coproc>
<file_ref>
<file_name>setiathome-r195.x86_64-pc-linux-gnu__cuda_fermi</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>

I think I have looked at it too long and am probably missing something obvious.

Any ideas?

Cheers

Mark

[arkayn]

I do believe that you are running into the mismatched library files problem. The x34 branch was CUDA 3.2 if I remember correctly and your libraries are now 4.0 and that is causing the segmentation violation.

Is it possible to install an older driver on the system?

Try this build as well..

[ausymark]

ok I tried that update after putting the *so.3 libs in the usr/lib64 , usr/lib64/nvidia-current and the seti@home project directory. Changed the app_info file to even directly point to the *so.3 files and had the same result as before. Removed the direct references to the *.so.3 files in app_info - still the same result.

I then removed all references to *so.3 to see if that might help - but no.

I guess a pertinent question maybe if anyone is beta testing one compiled for CUDA 4?  (Unless you may have other ideas?)

Cheers

Mark

[arkayn]

We are not even doing much testing of it on Windows currently, too many problems with it at this early stage.

http://www.nvidia.com/object/linux-display-amd64-275.21-driver.html

That should work on your system, I hope. The 280 and up are 4.0 version drivers.

[ausymark]

That did it

Though I am running with one of the earlier CUDA seti versions, i will retry the beta ones you sent above sometime on the weekend. (I'm just glad its all working again lol). I will let you know which out of the two I got to run. (Which one do your think is the one that I should test out of the two?)

BTW I am interested in trying out Beta's to give feedback.

And thanks for the help I really appreciate it

Cheers

Mark

[ausymark]

OK, I have gotten the "setiathome-r195.x86_64-pc-linux-gnu__cuda_fermi" CUDA Beta executable with no issues so far.

Though is the "setiathome_x36k_x86_64-pc-linux-gnu_cuda32" a later beta? Should I be using it instead?

In hindsight I think the three main issues I had was that:

* The original nVidia driver was the latest - and hence Cuda V4
* I was unsuccessfully running the Cuda 4 libraries that Seti couldn't see
* The GPU within the Intel i7 2600K CPU was most likely confusing matters for the entire CUDA setup.

I hope others find the above useful :-)

Cheers

Mark

[arkayn]

r195 is a beta app, x36k was a alpha test version.

[ausymark]

Tried the alpha, x36, it just ripped through a whole set of CUDA work units with computation errors, I have gone back to the beta, r195, for the moment.

Cheers

Mark

[arkayn]

Of course, now you are a member of Lunatics again as well.

[ausymark]

r195 is a beta app, x36k was a alpha test version.

Hi Arkavn

Do we have any new Alphas/Betas for the Seti nvidia Fermi CUDA 3's?

For some strange reason my PC has started crashing, and rolling back to the original Fermi code hasnt helped (I am also testing several other hardware related stuff)

Cheers

Mark

[arkayn]

Unfortunately, you have every build that I have access too.

Aaron did just come back to Lunatics, so he might get another build out eventually.

[ausymark]

ok

we will see what happens.

thx for getting back to me so quickly

cheers

Mark

[arkayn]

Looks like they just released the x36k version into the wild, you might want to post your findings on that one over there.