Linux CUDA 'Special' App finally available, featuring Low CPU use

Message boards : Number crunching : Linux CUDA 'Special' App finally available, featuring Low CPU use
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 62 · 63 · 64 · 65 · 66 · 67 · 68 . . . 83 · Next

AuthorMessage
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1895829 - Posted: 17 Oct 2017, 8:56:16 UTC - in response to Message 1895826.  
Last modified: 17 Oct 2017, 8:56:46 UTC

It appears you're getting the False Overflows. Try removing that setting: Using pfb = 32 from command line args https://setiathome.berkeley.edu/result.php?resultid=6097878442
If you see a couple more switch to the other Apps.
ID: 1895829 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895830 - Posted: 17 Oct 2017, 9:11:55 UTC - in response to Message 1895829.  

It appears you're getting the False Overflows. Try removing that setting: Using pfb = 32 from command line args https://setiathome.berkeley.edu/result.php?resultid=6097878442
If you see a couple more switch to the other Apps.


. . OK now I am confused again, I removed -nobs and -pfb 32 at the same time from app_info.xml

. . Where else can it be getting that from ???

Stephen

??
ID: 1895830 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895834 - Posted: 17 Oct 2017, 9:56:40 UTC - in response to Message 1895829.  

. . OK I have looked at the first 5 tasks that validated and even the sixth and they all show -pfb of 32. The first five have ARs of about 0.7 so they were "optimistic" run times, the sixth is AR of 0.41 so is "normal". Thankfully none of them are overflows.

. . manager was not running when I edited app_info.xml BUT, the client starts at boot and when I exited manager I may not have closed the client ....

. . I have checked app_info.xml and the changes are definitely there so I have closed Manager and restarted the client, will have to wait for more results to confirm that -pfb=32 is gone.

. . Then comes the scary part, editing app_info.xml for the new app ...

Stephen

?/
ID: 1895834 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895844 - Posted: 17 Oct 2017, 11:55:25 UTC

. . An observation, the post processing delay on this card is horrendous. The app completed at just under 23 mins but the counter continued for another 2.5 mins before the status changed to uploading. Now that seems just wrong to me :)

Stephen

??
ID: 1895844 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895873 - Posted: 18 Oct 2017, 1:02:03 UTC - in response to Message 1895834.  


. . I have checked app_info.xml and the changes are definitely there so I have closed Manager and restarted the client, will have to wait for more results to confirm that -pfb=32 is gone.
. . Then comes the scary part, editing app_info.xml for the new app ...
Stephen


. . Well that got rid of -pfb=32. Since then a few jobs have been overflows but so far (touch wood) nothing has failed to validate. The single digit inconclusives are with -pfb=32. Hopefully they will yet be vindicated.

. . I think perhaps you have enough results for a comparison?

. . So what's the good oil on editing app_info.xml for the new app??

Stephen

??
ID: 1895873 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895874 - Posted: 18 Oct 2017, 1:18:05 UTC - in response to Message 1895834.  
Last modified: 18 Oct 2017, 1:32:16 UTC


. . I have checked app_info.xml and the changes are definitely there so I have closed Manager and restarted the client, will have to wait for more results to confirm that -pfb=32 is gone.
. . Then comes the scary part, editing app_info.xml for the new app ...
Stephen


. . Well that got rid of -pfb=32. Since then a few jobs have been overflows but so far (touch wood) nothing has failed to validate. The single digit inconclusives are with -pfb=32. Hopefully they will yet be vindicated.

. . I think perhaps you have enough results for a comparison?

. . So what's the good oil on editing app_info.xml for the new app?? I can see the two lines that refer to the app itself and 4 lines that refer to the 2 associated libraries. It is straightforward what the new app name should be but what names do I use for the 2 libraries??

Stephen

??
ID: 1895874 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895889 - Posted: 18 Oct 2017, 2:50:25 UTC

. . @ TBar

. . UPDATE:

. . The first job has failed to validate. It was one of the very first tasks run and has -pfb=32.

http://setiathome.berkeley.edu/results.php?hostid=8222433&offset=0&show_names=0&state=5&appid=

Stephen

:(
ID: 1895889 · Report as offensive
Bruce
Volunteer tester

Send message
Joined: 15 Mar 02
Posts: 123
Credit: 124,955,234
RAC: 11
United States
Message 1895892 - Posted: 18 Oct 2017, 3:22:41 UTC

TBar,
I just tried that xs2-35_cuda90 and all the ones I ran just ended in comp error. Tried both of the ones you posted.
Have now switched back to the 3t2b_cuda65 app.
Kind of bummed at the moment.
Just have to keep trying, I guess.
Bruce
ID: 1895892 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1895895 - Posted: 18 Oct 2017, 4:16:14 UTC - in response to Message 1895892.  
Last modified: 18 Oct 2017, 5:00:34 UTC

That's interesting, that's the same error I get when you try to compile the App with sm=30. But your cards are supposed to be sm=35. Those other Apps that work are compiled with sm=32 and up, maybe it would work using the CUDA 9 app with sm=32. I suppose it's worth a try. The New sm=32 App works in my machine, https://setiathome.berkeley.edu/result.php?resultid=6099460046 Check your PMs.

Stephen, I've already posted the instructions for changing Apps. The Library names should also be self evident, just look at the names that are already there. You have a pair of Libraries, one pair goes at the top, and one pair goes at the bottom. The CUDA 9 App has them built in, it should work by just changing the App names for the CUDA 9 App. It worked on my machine. Just remember to suspend all the tasks and resume just One to test the App. Try the CUDA 9 App next, see if you also get that error with the sm=35 App.
ID: 1895895 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895914 - Posted: 18 Oct 2017, 10:30:28 UTC - in response to Message 1895895.  


Stephen, I've already posted the instructions for changing Apps. The Library names should also be self evident, just look at the names that are already there. You have a pair of Libraries, one pair goes at the top, and one pair goes at the bottom. The CUDA 9 App has them built in, it should work by just changing the App names for the CUDA 9 App. It worked on my machine. Just remember to suspend all the tasks and resume just One to test the App. Try the CUDA 9 App next, see if you also get that error with the sm=35 App.


. . So if leave the libraries as they are for CUDA80 the CUDA90 app won't mind? It scares me but I'll try anything once ...

Stephen

???
ID: 1895914 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1895918 - Posted: 18 Oct 2017, 11:11:49 UTC - in response to Message 1895895.  
Last modified: 18 Oct 2017, 11:24:12 UTC

That's interesting, that's the same error I get when you try to compile the App with sm=30. But your cards are supposed to be sm=35. Those other Apps that work are compiled with sm=32 and up, maybe it would work using the CUDA 9 app with sm=32. I suppose it's worth a try. The New sm=32 App works in my machine, https://setiathome.berkeley.edu/result.php?resultid=6099460046 Check your PMs.


. . Hi TBar

. . Well there are 3 invalids now, 2 from the first batch of about 2 dozen tasks where -pfb=32 was set and just one so far from after it was cleared. Since there have been more than 50 tasks completed so far that seems not as bad as it might be. But I am about to make the change over to 3xs3-35. So far nothing new in my pm-inbox.

. . {edit} uh-oh! Instant computation error :( Now in limbo waiting for next step ...

Stephen

..
ID: 1895918 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1895960 - Posted: 18 Oct 2017, 15:27:22 UTC - in response to Message 1895918.  
Last modified: 18 Oct 2017, 15:28:28 UTC

So, yours gives the same error. The 32 App probably won't work either, but, you might as well try it. Just change the 35 in the link to 32 and the same link should work. If the 32 App doesn't work give the CUDA65 App a try. I'll see about making a non-static zi3x later and see if that works on the cc 3.5 cards, those old cards seem to have reached their limit.
ID: 1895960 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1896008 - Posted: 18 Oct 2017, 20:48:24 UTC - in response to Message 1895960.  

. . Same thing, I tired 2 tasks this time to be sure etc.

Stephen

:(
ID: 1896008 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1896011 - Posted: 18 Oct 2017, 20:50:41 UTC - in response to Message 1895960.  

So, yours gives the same error. The 32 App probably won't work either, but, you might as well try it. Just change the 35 in the link to 32 and the same link should work. If the 32 App doesn't work give the CUDA65 App a try. I'll see about making a non-static zi3x later and see if that works on the cc 3.5 cards, those old cards seem to have reached their limit.


. . Since the fan control script doesn't work with the 384.90 driver I am wondering if the video driver doesn't like these cards ...

Stephen

??
ID: 1896011 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1896026 - Posted: 18 Oct 2017, 21:29:01 UTC - in response to Message 1896011.  

When I updated to the 387.12 driver, I did have to go back through the coolbits tweak and reset the Nvidia X-server settings. But I still have Jeff's fan control app working just as before. I wouldn't think that the fan control script works any differently. Still the same mechanism under the covers.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1896026 · Report as offensive
Profile Jeff Buck Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1896030 - Posted: 18 Oct 2017, 21:37:55 UTC - in response to Message 1896011.  

. . Since the fan control script doesn't work with the 384.90 driver
What error message are you getting?

Are you using "GPUTargetFanSpeed" (for newer drivers) or "GPUCurrentFanSpeed (for older drivers)?
ID: 1896030 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1896031 - Posted: 18 Oct 2017, 21:40:46 UTC - in response to Message 1896026.  

When I updated to the 387.12 driver, I did have to go back through the coolbits tweak and reset the Nvidia X-server settings. But I still have Jeff's fan control app working just as before. I wouldn't think that the fan control script works any differently. Still the same mechanism under the covers.


. . For some reason it bombs out badly on my rig since upgrading to 384.90. Like the commands are failing to be recognised by the driver/xserver.

Stephen

??
ID: 1896031 · Report as offensive
Profile Jeff Buck Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1896036 - Posted: 18 Oct 2017, 22:11:19 UTC - in response to Message 1896026.  

When I updated to the 387.12 driver, I did have to go back through the coolbits tweak and reset the Nvidia X-server settings. But I still have Jeff's fan control app working just as before. I wouldn't think that the fan control script works any differently. Still the same mechanism under the covers.
One of the improvements in the GUI, besides it being, well, a GUI, is that it should work with both the older and newer drivers without any internal changes. Right up front it checks to see if "GPUTargetFanSpeed" is a valid attribute for the driver in use. If it gets an error, it falls back to "GPUTargetFanSpeed" and continues on its merry way.
ID: 1896036 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1896040 - Posted: 18 Oct 2017, 22:36:31 UTC - in response to Message 1896036.  
Last modified: 18 Oct 2017, 22:39:14 UTC

One of the improvements in the GUI, besides it being, well, a GUI, is that it should work with both the older and newer drivers without any internal changes. Right up front it checks to see if "GPUTargetFanSpeed" is a valid attribute for the driver in use. If it gets an error, it falls back to "GPUTargetFanSpeed" and continues on its merry way.

. . I replaced GPUCurrentFanSpeed with GPUTargetFanSpeed and it still bombs ... :( Maybe I will have to get the GUI version set up. Now where did I put that download?

Stephen

??
ID: 1896040 · Report as offensive
Profile Jeff Buck Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1896042 - Posted: 18 Oct 2017, 22:51:48 UTC - in response to Message 1896040.  
Last modified: 18 Oct 2017, 23:28:30 UTC

. . I replaced GPUCurrentFanSpeed with GPUTargetFanSpeed and it still bombs ... :( Maybe I will have to get the GUI version set up. Now where did I put that download?

Stephen

??
Without knowing what error message(s) you're getting, it's impossible to even attempt a diagnosis. Have you tried using the GUI Fan Control yet?

EDIT: Scratch that last question since I see you answered it before I asked it. I guess I didn't read past "it still bombs". ;^)
ID: 1896042 · Report as offensive
Previous · 1 . . . 62 · 63 · 64 · 65 · 66 · 67 · 68 . . . 83 · Next

Message boards : Number crunching : Linux CUDA 'Special' App finally available, featuring Low CPU use


 
©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.