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 . . . 76 · 77 · 78 · 79 · 80 · 81 · 82 . . . 83 · Next

AuthorMessage
rob smith Special Project $250 donor
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 16595
Credit: 346,366,099
RAC: 230,937
United Kingdom
Message 1913960 - Posted: 19 Jan 2018, 8:26:26 UTC

Setting a small (5-10) startup delay can be very useful as it allows the video drivers a chance to settle down before they get kicked by BOINC then SETI which can result in them throwing a wobbly.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1913960 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 5910
Credit: 415,157,734
RAC: 1,063,421
United States
Message 1913969 - Posted: 19 Jan 2018, 9:32:14 UTC

TBar, is that installation using the Wayland desktop manager? Do you have any issues running Nvidia drivers? Did you use that
xhost si:localuser:root
command line trick to fix the X.org user authorization problem for graphical apps?
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1913969 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 4435
Credit: 285,138,943
RAC: 507,652
United States
Message 1913993 - Posted: 19 Jan 2018, 12:49:55 UTC - in response to Message 1913969.  

I don't know about any Wayland, and since BOINC is running under the user account there is absolutely no problems with any sort of authorization. Lubuntu uses the exact same GPU drivers as Ubuntu, I don't know of any trouble with Lubuntu, and the run times are almost identical to those when running Ubuntu 17.10. The reason for Lubuntu is to provide a lightweight OS that works well with 'lightweight' hardware. It has an XP like Task Bar and start Menu which should make it a big hit with all those XP & Win 7 holdovers 'cause we know from Win 8 that all 90% of Windoze users care about is their Start Menu.



You can read about it here, https://wiki.ubuntu.com/Lubuntu
Hmmm, an XP Start Menu with the ability to run the latest nVidia GPUs, what more could an XP fanboy wish for?

;-)
ID: 1913993 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 5910
Credit: 415,157,734
RAC: 1,063,421
United States
Message 1914044 - Posted: 19 Jan 2018, 18:34:50 UTC - in response to Message 1913993.  

OK, thanks. I am still very confused. All that I have read is that Artful Aardvark (17.04/10) ships with the Wayland display manager as default instead of X11. Wayland is supposed to be the default going forward with the Gnome desktop. Supposedly at Login, you have a choice of either booting the default display manager (Wayland) or (X.server under Wayland) There is supposed to be no compatibility with closed source proprietary drivers like Nvidia. Wayland is just supposed to work with Intel, Radeon and Nouveau kernel drivers.

Why I asked if you had any troubles with the Nvidia drivers. From your post .... guess not.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1914044 · Report as offensive
Stephen "Heretic" Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 3779
Credit: 89,137,393
RAC: 161,852
Australia
Message 1914059 - Posted: 19 Jan 2018, 19:32:30 UTC - in response to Message 1913993.  

I don't know about any Wayland, and since BOINC is running under the user account there is absolutely no problems with any sort of authorization. Lubuntu uses the exact same GPU drivers as Ubuntu, I don't know of any trouble with Lubuntu, and the run times are almost identical to those when running Ubuntu 17.10. The reason for Lubuntu is to provide a lightweight OS that works well with 'lightweight' hardware. It has an XP like Task Bar and start Menu which should make it a big hit with all those XP & Win 7 holdovers 'cause we know from Win 8 that all 90% of Windoze users care about is their Start Menu.



You can read about it here, https://wiki.ubuntu.com/Lubuntu
Hmmm, an XP Start Menu with the ability to run the latest nVidia GPUs, what more could an XP fanboy wish for?

;-)


. . I'll drink to that! Finally, a Linux incarnation that is civilised :)

Stephen

:)
ID: 1914059 · Report as offensive
klepel

Send message
Joined: 8 Jun 10
Posts: 11
Credit: 8,214,764
RAC: 7,255
Peru
Message 1914076 - Posted: 19 Jan 2018, 20:28:50 UTC

@TBar: If the file WU name, which shows-up in BOINC Manager, is: "setiathome_v8 8.01 (cuda90)", and the expected run times is about 2.5 times less than with SoGs. Then everything is set-up and running!

Thanks a lot for your help!
ID: 1914076 · Report as offensive
Stephen "Heretic" Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 3779
Credit: 89,137,393
RAC: 161,852
Australia
Message 1914077 - Posted: 19 Jan 2018, 20:31:46 UTC
Last modified: 19 Jan 2018, 20:34:04 UTC

. . I find this curious, I have noticed a large number of Blc02 tasks that are taking 10 times as long to complete as the majority of Blc tasks do. Instead of 3 mins ( on the GTX 970s) they are taking 30.

http://setiathome.berkeley.edu/result.php?resultid=6325970436

http://setiathome.berkeley.edu/result.php?resultid=6325970436

http://setiathome.berkeley.edu/result.php?resultid=6325902136

. . I am seeing dozens and dozens of them, maybe one task in 20 or 30. Anyone have any idea what the difference is? They are similar ARs to other Blc tasks that are taking just 2 mins.

. . They are not happening on my Windows box nor on the other Linux box. Just the rig with the 970s. Clearly something is wrong and I would like to be able to fix it.

Stephen

??
ID: 1914077 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 5910
Credit: 415,157,734
RAC: 1,063,421
United States
Message 1914080 - Posted: 19 Jan 2018, 20:41:22 UTC - in response to Message 1914077.  

Stephen, those are CPU tasks. They are supposed to take that long.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1914080 · Report as offensive
Stephen "Heretic" Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 3779
Credit: 89,137,393
RAC: 161,852
Australia
Message 1914117 - Posted: 20 Jan 2018, 0:22:25 UTC - in response to Message 1914080.  
Last modified: 20 Jan 2018, 0:28:23 UTC

Stephen, those are CPU tasks. They are supposed to take that long.


. . D'oh! The things you miss when you are half asleep ... :(

. . Excuse me while I crawl back into the corner and have a nap :(

. . To put it into perspective though, I was looking at the stats from my Windows rig and the other Linux rig doesn't do any CPU crunching, and for that period of time I completely forgot that it's big brother does ... no excuse of course ... But the pattern should have made me think, one in twenty is the ratio of CPU to GPU tasks ... I think I need a holiday. Just consider me the light relief ...

Stephen

Oopss!
ID: 1914117 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1675
Credit: 388,101,452
RAC: 296,340
United States
Message 1914119 - Posted: 20 Jan 2018, 0:29:45 UTC - in response to Message 1914059.  

. . I'll drink to that! Finally, a Linux incarnation that is civilised :)

Stephen

:)

+1 from me!

ID: 1914119 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 5910
Credit: 415,157,734
RAC: 1,063,421
United States
Message 1914123 - Posted: 20 Jan 2018, 0:45:19 UTC - in response to Message 1914117.  

Stephen, those are CPU tasks. They are supposed to take that long.


. . D'oh! The things you miss when you are half asleep ... :(

. . Excuse me while I crawl back into the corner and have a nap :(

. . To put it into perspective though, I was looking at the stats from my Windows rig and the other Linux rig doesn't do any CPU crunching, and for that period of time I completely forgot that it's big brother does ... no excuse of course ... But the pattern should have made me think, one in twenty is the ratio of CPU to GPU tasks ... I think I need a holiday. Just consider me the light relief ...

Stephen

Oopss!

Yep, that was a grinner .... thanks for the comedic relief.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1914123 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 4435
Credit: 285,138,943
RAC: 507,652
United States
Message 1914246 - Posted: 20 Jan 2018, 17:22:36 UTC - in response to Message 1914044.  

OK, thanks. I am still very confused. All that I have read is that Artful Aardvark (17.04/10) ships with the Wayland display manager as default instead of X11. Wayland is supposed to be the default going forward with the Gnome desktop. Supposedly at Login, you have a choice of either booting the default display manager (Wayland) or (X.server under Wayland) There is supposed to be no compatibility with closed source proprietary drivers like Nvidia. Wayland is just supposed to work with Intel, Radeon and Nouveau kernel drivers.

Why I asked if you had any troubles with the Nvidia drivers. From your post .... guess not.

Well, the Manager in Lubuntu 17.10.1 is still lightdm, you can find it in etc/X11/default-display-manager. I booted back into Ubuntu 17.10 and it is listed as gdm3 there. I didn't have any trouble installing nVidia driver 390.12 from nVidia using the recovery mode, and it seems to be running a little faster than the 384.111 driver, https://setiathome.berkeley.edu/show_host_detail.php?hostid=6906726 It seems to be impossible to get auto login to work, but I don't see any choices there. It just shows my name and I have to enter the password. Other than the login everything seems to be the same. It does seem easier to install the driver from recovery mode now, as trying to drop into the console with alt+Ctrl+F3 doesn't work very well.
I dunno, doesn't seem to be much different to me.
ID: 1914246 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 5910
Credit: 415,157,734
RAC: 1,063,421
United States
Message 1914251 - Posted: 20 Jan 2018, 17:45:23 UTC - in response to Message 1914246.  

Good to get a report from you, TBar. I guess all my reading about the troubles in the transition from X11 and Unity to Gnome aren't coming to pass. I was just scared of how much upset we might have using Nvidia drivers come April when Bionic Beaver 18.04 LTS drops and the switch away from lightdm and Unity to Wayland and Gnome.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1914251 · Report as offensive
Profile Jeff Buck Special Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1914682 - Posted: 23 Jan 2018, 3:56:35 UTC - in response to Message 1912865.  

I don't get upset about Petri trying out the latest zi3xs4 on Main. It is just one computer. Now if he made that app available to the public and dozens of computers were trying to run it, I would have an issue.
Well, although zi3xs4 may not be "publicly" available, I just noticed that Laurent (W3Perl) is running it on some of his machines. I don't know how many of them, because I stopped counting after 6. It's probably a safe bet that there are more. "Dozens" might very well be apt.

So.......somewhere in there, I would also say it's a safe bet that there have been cross-validations between Petri's box and Laurent's. If any of those involve that 30-Pulse overflow bug that Petri was testing a fix for last week, then that's definitely bad, because they would have either overruled a 3rd host reporting legitimate results or, if they were the first two hosts on the WU, prevented a non-impaired host from even getting that WU. That situation should be completely unacceptable.
ID: 1914682 · Report as offensive
Stephen "Heretic" Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 3779
Credit: 89,137,393
RAC: 161,852
Australia
Message 1914703 - Posted: 23 Jan 2018, 8:09:02 UTC - in response to Message 1914682.  

I don't get upset about Petri trying out the latest zi3xs4 on Main. It is just one computer. Now if he made that app available to the public and dozens of computers were trying to run it, I would have an issue.
Well, although zi3xs4 may not be "publicly" available, I just noticed that Laurent (W3Perl) is running it on some of his machines. I don't know how many of them, because I stopped counting after 6. It's probably a safe bet that there are more. "Dozens" might very well be apt.

So.......somewhere in there, I would also say it's a safe bet that there have been cross-validations between Petri's box and Laurent's. If any of those involve that 30-Pulse overflow bug that Petri was testing a fix for last week, then that's definitely bad, because they would have either overruled a 3rd host reporting legitimate results or, if they were the first two hosts on the WU, prevented a non-impaired host from even getting that WU. That situation should be completely unacceptable.


. .To be Frank [:)] for once I share your concern, it seems that version is still too early in development to be shared. But if it was to be anyone I believe Laurent is a very capable person. Perhaps on one box but not on so many of his horde of crunchers.

Stephen

? ?
ID: 1914703 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 5910
Credit: 415,157,734
RAC: 1,063,421
United States
Message 1914705 - Posted: 23 Jan 2018, 8:19:38 UTC - in response to Message 1914703.  

Well, we can't control Petri33 and who he might have running his latest alpha. I too am a bit alarmed that he has persuaded Laurent to be his testbed on so many machines.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1914705 · Report as offensive
Profile Jeff Buck Special Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1914902 - Posted: 24 Jan 2018, 20:17:05 UTC - in response to Message 1914682.  

I don't get upset about Petri trying out the latest zi3xs4 on Main. It is just one computer. Now if he made that app available to the public and dozens of computers were trying to run it, I would have an issue.
Well, although zi3xs4 may not be "publicly" available, I just noticed that Laurent (W3Perl) is running it on some of his machines. I don't know how many of them, because I stopped counting after 6. It's probably a safe bet that there are more. "Dozens" might very well be apt.

So.......somewhere in there, I would also say it's a safe bet that there have been cross-validations between Petri's box and Laurent's. If any of those involve that 30-Pulse overflow bug that Petri was testing a fix for last week, then that's definitely bad, because they would have either overruled a 3rd host reporting legitimate results or, if they were the first two hosts on the WU, prevented a non-impaired host from even getting that WU. That situation should be completely unacceptable.
Inasmuch as it would be impossible for me to manually go through all of Petri's or Laurent's tasks to see how often those 30-Pulse overflows might have cross-validated, I've instead used my own hosts' history with Petri as wingman to come up with an estimate.

In the last 30 days, Petri's host has been wingman on 319 WUs across my 3 Linux boxes. Excluding the 20 of his "ghosts" that have yet to time out, that leaves 299 tasks that have been completed. Of those, Petri's task has been marked as Invalid 6 times, most likely due to that 30-Pulse overflow bug. Had Petri's wingman for those tasks been one of Laurent's boxes running zi3xs4, they would likely have successfully validated, putting garbage into the DB.

Now, I don't know exactly how many of Laurent's boxes are running zi3xs4. though I did identify that 24 out of the first 34 that I looked at were running it, about 70%. Since Laurent's current RAC is about 10 times mine, I think it's probably safe to assume that he's processing about 10 times the number of tasks that I am, also. And if 70% of them are run with zi3xs4. I would estimate that Petri has been his wingman on about 2,100 of those zi3xs4 tasks. Therefore, if the ratio of Invalids that I saw from Petri's tasks against my hosts holds, a reasonable estimate is that somewhere around 42 of those bogus 30-Pulse overflows will have been cross-validated and found their way into the DB. That's just in the last 30 days.

In my opinion, that needs to stop immediately. If Petri's come up with a fix for that bug, and it appeared as if he did, based on the test results he posted last week, that fix needs to be implemented NOW. Either that, or get zi3xs4 out of the production environment entirely.
ID: 1914902 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 4435
Credit: 285,138,943
RAC: 507,652
United States
Message 1919894 - Posted: 19 Feb 2018, 15:58:53 UTC

Just a reminder for those running the CUDA Special App.

The last, and Best, working version zi3v is posted at Crunchers Anonymous. The 'BOINC All-In-One Build to run zi3v CUDA 8.0 in Ubuntu 12.04 to 17.04' download contains both a nicely working copy of BOINC 7.4.44 and the CUDA 8 version of the CUDA App. This package makes it easy to have BOINC up and running from your Home folder, with the Special App, on a New Ubuntu install in about 5 minutes...or less.

The other download 'Linux CUDA Special Apps' has links to the CUDA 9.0 version and the CUDA 6.0 version of zi3v to be installed (copied) into an existing setiathome.berkeley.edu folder, preferably in a BOINC folder in your Home folder.

Since the Arecibo work has stopped it has become clear Most of the previous Inconclusive Results were resulting from the Arecibo tasks. Now with just the BLC tasks the CUDA zi3v has Very Few Inconclusive results, mostly from Bad WingPeople and a couple of Flaws in other Apps. There is still an occasional Bad Best Pulse, but much fewer than the Flaws in the other Apps, most notably the Missing Pulse with a score of exactly One that occurs with Most the Apps compiled from the AKv8 source.

So Again, it is requested that ANYONE running an Older Version of the CUDA Special App upgrade to the zi3v version to reduce the number of Inconclusive Results being produced. In the cases I've seen the CUDA zi3v is Faster than those Older versions so it is a Win Win upgrade, the Task is completed in less time with Fewer Inconclusive results.

Upgrade to zi3v Today.
ID: 1919894 · Report as offensive
Stephen "Heretic" Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 3779
Credit: 89,137,393
RAC: 161,852
Australia
Message 1919966 - Posted: 19 Feb 2018, 22:28:11 UTC - in response to Message 1919894.  

Just a reminder for those running the CUDA Special App.

The last, and Best, working version zi3v is posted at Crunchers Anonymous. The 'BOINC All-In-One Build to run zi3v CUDA 8.0 in Ubuntu 12.04 to 17.04' download contains both a nicely working copy of BOINC 7.4.44 and the CUDA 8 version of the CUDA App. This package makes it easy to have BOINC up and running from your Home folder, with the Special App, on a New Ubuntu install in about 5 minutes...or less.

Since the Arecibo work has stopped it has become clear Most of the previous Inconclusive Results were resulting from the Arecibo tasks. Now with just the BLC tasks the CUDA zi3v has Very Few Inconclusive results, mostly from Bad WingPeople and a couple of Flaws in other Apps. There is still an occasional Bad Best Pulse, but much fewer than the Flaws in the other Apps, most notably the Missing Pulse with a score of exactly One that occurs with Most the Apps compiled from the AKv8 source.

So Again, it is requested that ANYONE running an Older Version of the CUDA Special App upgrade to the zi3v version to reduce the number of Inconclusive Results being produced. In the cases I've seen the CUDA zi3v is Faster than those Older versions so it is a Win Win upgrade, the Task is completed in less time with Fewer Inconclusive results.

Upgrade to zi3v Today.


. . +1 - 100% ... This version is working very, very well, the inconclusive rate is quite comparable to SoG and it is very consistent in performance.

. . But I have one question. It seems that you are recommending that people setting up a new cruncher opt for the Cuda 8.0 version rather than the Cuda 9.0 version, is there a reason for this?

Stephen

? ?
ID: 1919966 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 4435
Credit: 285,138,943
RAC: 507,652
United States
Message 1919992 - Posted: 19 Feb 2018, 23:35:28 UTC - in response to Message 1919966.  

Actually, if you look at the Post date you will see it was made before CUDA 9 was available. I see no reason to change it as any given repository will offer CUDA 8 support before CUDA 9.
ID: 1919992 · Report as offensive
Previous · 1 . . . 76 · 77 · 78 · 79 · 80 · 81 · 82 . . . 83 · Next

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


 
©2018 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.