V8 CUDA for Linux?

Message boards : Number crunching : V8 CUDA for Linux?
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 10 · Next

AuthorMessage
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1774823 - Posted: 28 Mar 2016, 21:27:38 UTC - in response to Message 1774334.  

More complex

Here all programs I see in my setiathome.berkeley.edu folder
/var/lib/boinc-client/projects/setiathome.berkeley.edu/astropulse_7.00_i686-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/astropulse_7.00_x86_64-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/astropulse_7.04_i686-pc-linux-gnu__sse
/var/lib/boinc-client/projects/setiathome.berkeley.edu/astropulse_7.04_x86_64-pc-linux-gnu__sse2
/var/lib/boinc-client/projects/setiathome.berkeley.edu/astropulse_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setiathome_7.01_i686-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setiathome_7.01_x86_64-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setiathome_7.08_x86_64-pc-linux-gnu__opencl_nvidia_sah
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setiathome_8.00_i686-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setiathome_8.00_x86_64-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setigraphics_7.01_i686-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setigraphics_7.01_x86_64-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setigraphics_8.00_i686-pc-linux-gnu
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setigraphics_8.00_x86_64-pc-linux-gnu

I Immagine I need an app_info entra for each one.

Correct?
ID: 1774823 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14679
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1774833 - Posted: 28 Mar 2016, 21:42:18 UTC - in response to Message 1774823.  

You only need to pick one of each type, ideally the 'best' (that's what the server was searching for).

I'd guess those would be

astropulse_7.04_x86_64-pc-linux-gnu__sse2
astropulse_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100
setiathome_8.00_x86_64-pc-linux-gnu

You may as well ignore all setiathome_7... - they're done now, or near enough.

All the setigraphics_ are screensavers to run alongside the other apps.

You may wish to refer to

SETI@Home Version 8 - app_info.xml files for upgrading Lunatics and anonymous platform (all OS)

while you work - the adaptations should be fairly obvious.
ID: 1774833 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1774854 - Posted: 29 Mar 2016, 0:17:30 UTC - in response to Message 1774823.  
Last modified: 29 Mar 2016, 0:36:47 UTC

If you look at the Downloaded file app_info.xml~ you will see the AP sections already there. All you have to do is add those sections to the Base app_info and make sure the file names match the Apps you intend to use. That's IF you want to run APs on that particular machine. As I already mentioned, the GTS 250s will produce More Credit running CUDA tasks than APs. I wouldn't even bother with APs on the GTS 250 machines.

The first thing you should do is make a New folder, call it Stock, and then copy and paste ALL the Stock Apps and associated files into it for safe keeping. When you switch from Stock to Anonymous platform the first thing the Server does is Delete ALL the Stock Apps and associated files from the setiathome.berkeley.edu folder. If you don't have another copy, you will have to track one down. After the Server deletes the files the first time, simply stop BOINC and paste the needed files back into the setiathome.berkeley.edu folder. They won't be deleted the second time.

IF you want to run the AP Apps, the app_info would be as such;

<app_info>
 <app>
    <name>astropulse_v7</name>
 </app>
  <file_info>
    <name>astropulse_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100</name>
    <executable/>
  </file_info>
  <file_info>
      <name>ap_cmdline_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100.txt</name>
  </file_info>
  <app_version>
    <app_name>astropulse_v7</app_name>
    <platform>x86_64-pc-linux-gnu</platform>
    <version_num>708</version_num>
    <plan_class>opencl_nvidia_100</plan_class>
    <coproc>
      <type>NVIDIA</type>
      <count>1</count>
    </coproc>
    <avg_ncpus>0.1</avg_ncpus>
    <max_ncpus>0.2</max_ncpus>
    <file_ref>
      <file_name>astropulse_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100</file_name>
      <main_program/>
    </file_ref>
    <file_ref>
       <file_name>ap_cmdline_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100.txt</file_name>
       <open_name>ap_cmdline.txt</open_name>
    </file_ref>
   </app_version>
  <app>
    <name>setiathome_v8</name>
  </app>
	<file_info>
	   <name>setiathome_x41zi_x86_64-pc-linux-gnu_cuda42</name>
	   <executable/>
	</file_info>
	<file_info>
	   <name>libcudart.so.4</name>
	    <executable/>
 	</file_info>
	<file_info>
	   <name>libcufft.so.4</name>
	   <executable/>
	</file_info>
	<app_version>
          <app_name>setiathome_v8</app_name>
	  <platform>x86_64-pc-linux-gnu</platform>
          <version_num>801</version_num>
	  <plan_class>cuda42</plan_class>
	    <avg_ncpus>0.05</avg_ncpus>
	    <max_ncpus>0.10</max_ncpus>
	     <coproc>
	       <type>CUDA</type>
	       <count>1.0</count>
	     </coproc>
	<file_ref>
           <file_name>setiathome_x41zi_x86_64-pc-linux-gnu_cuda42</file_name>
           <main_program/>
 	</file_ref>
	<file_ref>
	   <file_name>libcudart.so.4</file_name>
	</file_ref>
	<file_ref>
	   <file_name>libcufft.so.4</file_name>
	 </file_ref>
     </app_version>
  <app>
      <name>astropulse_v7</name>
  </app>
    <file_info>
        <name>astropulse_7.04_x86_64-pc-linux-gnu__sse2</name>
        <executable/>
    </file_info>
    <app_version>
        <app_name>astropulse_v7</app_name>
        <platform>x86_64-pc-linux-gnu</platform>
        <version_num>704</version_num>
        <file_ref>
            <file_name>astropulse_7.04_x86_64-pc-linux-gnu__sse2</file_name>
            <main_program/>
        </file_ref>
    </app_version>
  <app>
     <name>setiathome_v8</name>
  </app>
    <file_info>
      <name>MBv8_8.0r3305_ssse3_x86_64-pc-linux-gnu</name>
      <executable/>
    </file_info>
    <app_version>
      <app_name>setiathome_v8</app_name>
      <platform>x86_64-pc-linux-gnu</platform>
      <version_num>800</version_num>
      <file_ref>
        <file_name>MBv8_8.0r3305_ssse3_x86_64-pc-linux-gnu</file_name>
        <main_program/>
      </file_ref>
    </app_version>
</app_info>

Make sure you have a copy of AstroPulse_Kernels_r2750.cl or which ever one is already present in the setiathome.berkeley.edu when you try to run APs, the astropulse_7.08 App needs it to run. I wouldn't even bother with the Stock MB CPU App, the MBv8_8.0r3305 will probably be close to Twice as fast on your CPUs. I have very similar CPUs as yours, compare the times on these Hosts;
Intel 2.40GHz http://setiathome.berkeley.edu/results.php?hostid=7258715&offset=60
Intel 2.66GHz http://setiathome.berkeley.edu/results.php?hostid=7769537&offset=120

Oh look, an app_info that's not 5 pages long...how nice ;-)
ID: 1774854 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1774913 - Posted: 29 Mar 2016, 7:25:21 UTC
Last modified: 29 Mar 2016, 7:25:31 UTC

Thanks!
I will try this evening
ID: 1774913 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775577 - Posted: 1 Apr 2016, 16:01:05 UTC

Ok, ... running.
I will see results tomorrows.
http://setiathome.berkeley.edu/results.php?hostid=7866790&offset=0&show_names=0&state=0&appid=29

By the way:

"When you switch from Stock to Anonymous platform the first thing the Server does is Delete ALL the Stock Apps and associated files from the setiathome.berkeley.edu folder. If you don't have another copy, you will have to track one down. After the Server deletes the files the first time, simply stop BOINC and paste the needed files back into the setiathome.berkeley.edu folder. They won't be deleted the second time."


No, this did not happens. Old task still running.
The only thing I see is that "remaining (estimated) of old task is longer than before ... but it is decrementing very quickly.
ID: 1775577 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1775588 - Posted: 1 Apr 2016, 17:24:07 UTC - in response to Message 1775577.  
Last modified: 1 Apr 2016, 17:38:59 UTC

Hmmm, seems all the results are Overflows, which is highly unlikely. Checking your recent MBv7 results also shows Overflows. That particular machine may have been a bad choice for the first tests. I would suggest stopping BOINC and replacing the CUDA 42 App with the CUDA 60 App and seeing if it is any better. As long as you just replace the file names in the app_info and leave the version number & plan class the same you won't have any trouble. There is an example of an older CUDA 60 app in the app_info.xml~ file you downloaded. Basically you just change the App names and library names to match the new Version 8 CUDA 60 App.
ID: 1775588 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775593 - Posted: 1 Apr 2016, 18:16:29 UTC - in response to Message 1775588.  
Last modified: 1 Apr 2016, 18:21:56 UTC

Hmmm, seems all the results are Overflows, which is highly unlikely. Checking your recent MBv7 results also shows Overflows. That particular machine may have been a bad choice for the first tests. I would suggest stopping BOINC and replacing the CUDA 42 App with the CUDA 60 App and seeing if it is any better. As long as you just replace the file names in the app_info and leave the version number & plan class the same you won't have any trouble. There is an example of an older CUDA 60 app in the app_info.xml~ file you downloaded. Basically you just change the App names and library names to match the new Version 8 CUDA 60 App.

I don't see it!
I have made the change e few hours ago (3) and I see only one inconclusive result and 25 waiting for validation
---
http://setiathome.berkeley.edu/result.php?resultid=4829949197

As I see that I have (storically) 4 inconclusive and 75 valid, rate 1 to 25 is better. May be it's better if I switch off AP.
ID: 1775593 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1775598 - Posted: 1 Apr 2016, 18:31:24 UTC - in response to Message 1775593.  

Check the results, everyone I've seen is showing;
SETI@Home Informational message -9 result_overflow
NOTE: The number of results detected equals the storage space allocated.

http://setiathome.berkeley.edu/result.php?resultid=4829948705
To have every result an Overflow is Not normal. Most of those will probably be invalid. The most common reason for Incorrect Overflows is Overheating and Overclocking. If that isn't the case, the reason may be more complex.
I've been watching your MBv7 OpenCL results all turn to Invalids as well. They also show Overflows; http://setiathome.berkeley.edu/result.php?resultid=4752782814
ID: 1775598 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775615 - Posted: 1 Apr 2016, 19:19:57 UTC - in response to Message 1775598.  
Last modified: 1 Apr 2016, 20:00:12 UTC

Seen.
I'm not overclocking and temp is fine (GPU 54 C)
CPUs are water cooled (60 ~ 65 C)
ID: 1775615 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1775627 - Posted: 1 Apr 2016, 20:30:56 UTC - in response to Message 1775615.  

Probably the easiest way to determine if the card is responsible would be to swap the 560 with the 640. That way the same driver would work with the 640 and you probably wouldn't have to change any settings. It should work with a simple card swap. You can try the CUDA 60 App, but it probably won't help considering you're having the same problem with the OpenCL and CUDA 42 Apps.
ID: 1775627 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775628 - Posted: 1 Apr 2016, 20:34:14 UTC

I see I have 4 inconclusive tasks as SETI@home v8 Anonymous platform (NVIDIA GPU) but other hosts of the work units are inconclusive too with
SETI@home v8 v8.00 (cuda50) windows_intelx86,
SETI@home v8 v8.00 windows_intelx86,
SETI@home v8 v8.00 (opencl_ati5_nocal) windows_intelx86
ID: 1775628 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775629 - Posted: 1 Apr 2016, 20:37:01 UTC - in response to Message 1775627.  
Last modified: 1 Apr 2016, 20:39:05 UTC

Probably the easiest way to determine if the card is responsible would be to swap the 560 with the 640. That way the same driver would work with the 640 and you probably wouldn't have to change any settings. It should work with a simple card swap. You can try the CUDA 60 App, but it probably won't help considering you're having the same problem with the OpenCL and CUDA 42 Apps.


I prefer to activate v8 nvidia on an other linux system, where I have NVIDIA GeForce GTX 260.
But tomorrow evening.
Or I can activate v8 nvidia on the host in wich I have 640.
ID: 1775629 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1775647 - Posted: 1 Apr 2016, 21:19:40 UTC - in response to Message 1775629.  

In the meantime, most of the tasks completed with the 560 will eventually become Invalid. If you're lucky the real Overflows will eventually validate. I'd stop downloading new tasks for it. If you swapped the cards and it worked, then the rest of the tasks would validate.
ID: 1775647 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775653 - Posted: 1 Apr 2016, 22:05:18 UTC - in response to Message 1775647.  

I have switched from 352 to 340 driver.
Maybe the -9 overflow error will disappear
ID: 1775653 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1775675 - Posted: 2 Apr 2016, 0:14:29 UTC - in response to Message 1775653.  

Now with the different driver it seems BOINC is having a problem finding the card. Strange it didn't have that problem before.
I also looked up the stock clock for a 560, this page puts it at 1620 MHz https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_500_Series
Yours is listed as clockRate = 1850 MHz. Perhaps it's a factory overclock?
I'm not sure about how to lower the NV GPU clock in Ubuntu, but it might work better with it set a little lower.
ID: 1775675 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775708 - Posted: 2 Apr 2016, 6:13:52 UTC - in response to Message 1775675.  
Last modified: 2 Apr 2016, 6:32:13 UTC

This is my son's host, that now is in Australia.
He is a gamer so maybe he made something to the GPU clock or he bought a high-performance card.
I will ask him.
I don't know how to restore default clock in a Linux Mint system.
Sysinfo confirms 1851 clock.

Coming to results:

1) I see that now I have 44 inconclusive. Was 4 yesterday, before the change, so 40 are connected to CUDA42.
In all cases I see that (of course) the other hosts in work unit are "completed, validation inconclusive" too. I can't know if errors is mine or from other hosts but I suppose are mine. -9 overflow even with 340 driver.

2) zero errors, zero invalid

3) 3 valid results, even if -9 overflow. As I started abut 15 hours ago, it's normal that I get only 3 valid results. Better to wait 3 days, I think.

I tried to do same job on the host with 640 but ... I don't find nvidia-modprobe in synaptic package manager of that host. Very strange. Host are same linux (mint 17.3, cinnamon 64) and same repository. Any idea?

In any case I'm sure that all my other linux hosts with nvidia GPU are not overclocked.
ID: 1775708 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775709 - Posted: 2 Apr 2016, 6:54:29 UTC - in response to Message 1775675.  

Now with the different driver it seems BOINC is having a problem finding the card. Strange it didn't have that problem before.

Maybe this is because now I'm using 340 driver but nvidia-modprobe is still 352.79

nvidia-modprobe is a simple setuid root utility which loads the NVIDIA kenel
module and creates the NVIDIA device files, to allow usage of the NVIDIA
driver by client applications
that would otherwise not have sufficient
permission to load the module or create the device files on their own.


Maybe I have to find 340.96 nvidia-modprobe or to revert to 352 driver.
I do (now) this second option.
ID: 1775709 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1775711 - Posted: 2 Apr 2016, 7:39:49 UTC - in response to Message 1775709.  
Last modified: 2 Apr 2016, 7:46:02 UTC

Here's instructions on how to set the Clock in Mint;
https://mintguide.org/system/445-overclocking-the-nvidia-graphics-card-on-linux-mint.html
I would change the settings back close to Stock or try Option 4. Perhaps Unchecking the box "Enable Overclocking" would suffice.

1. Open a terminal, copy and run the following commands:
sudo nvidia-xconfig
sudo nvidia-xconfig --cool-bits=5

2. Then restart the computer.

3. After a new login open from the system menu NVIDIA X Server Settings and go to Thermal Settings, where you should now see the advanced settings control the rotation speed of the cooling fan on the video card.

4. If something goes wrong, you can always reset thermal settings by clicking Reset to Hardware Default.

Note.

cool-bits=1 enables the possibility of overclocking
cool-bits=4 includes the ability to manually control the fan
cool-bits=5 includes both



Here's another guide; http://www.phoronix.com/scan.php?px=MTY1OTM&page=news_item
ID: 1775711 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775714 - Posted: 2 Apr 2016, 8:25:36 UTC - in response to Message 1775711.  
Last modified: 2 Apr 2016, 8:30:36 UTC

done
now, after a strange series for reboots, I have the possibility to change fan speed but I don't see the clock frequency tab.
My son is telling me now that the board was already overclocked.
I will look for the box.

Found: factory overclocked
ID: 1775714 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1775787 - Posted: 2 Apr 2016, 18:49:30 UTC

summary

all my free GPUs are now working with SETIv8_Linux_CUDA42.7z

NVIDIA GeForce GTX 260
NVIDIA GeForce GT 640
NVIDIA GeForce GTX 560
NVIDIA GeForce GTS 250

All is working fine but GeForce GTX 560, perhaps because it's factory overclocked.
ID: 1775787 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 10 · Next

Message boards : Number crunching : V8 CUDA for Linux?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.