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 . . . 79 · 80 · 81 · 82 · 83 · Next

AuthorMessage
Profile [AF>EDLS]GuL
Volunteer tester

Send message
Joined: 15 Feb 06
Posts: 10
Credit: 27,125,503
RAC: 0
France
Message 1930463 - Posted: 17 Apr 2018, 6:54:18 UTC

Hi all,
I have now a GTX 1060 6Go and a GTX 1070 using zi3v-Cuda9.0 under Linux Fedora 27. However, the computation is not always using GPU, sometimes down to 70 %. Is there a way to improve this ? Can I run two tasks in parallel on each GPU or wil it give unreliable results?

For information, I have at this time 0 error, 0 invalid, 24 inconclusive, 626 valid and 987 pending tasks.
ID: 1930463 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22492
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1930465 - Posted: 17 Apr 2018, 7:06:44 UTC

The application is not designed to support two tasks per GPU. It may work, but it may result in very much extended run times (over those for one task per GPU)
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1930465 · Report as offensive
Profile [AF>EDLS]GuL
Volunteer tester

Send message
Joined: 15 Feb 06
Posts: 10
Credit: 27,125,503
RAC: 0
France
Message 1930468 - Posted: 17 Apr 2018, 7:13:51 UTC - in response to Message 1930465.  

Ok, thanks. Can I had command line options to increase GPU occupation ?
ID: 1930468 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1930480 - Posted: 17 Apr 2018, 8:20:01 UTC - in response to Message 1930468.  

-nobs forces 100% CPU use, which makes sure the GPU is not waiting.
Also don't pin your CPU usage. You will want 3 or 4 free threads for the 2 GPUs.
ID: 1930480 · Report as offensive
Profile [AF>EDLS]GuL
Volunteer tester

Send message
Joined: 15 Feb 06
Posts: 10
Credit: 27,125,503
RAC: 0
France
Message 1930483 - Posted: 17 Apr 2018, 8:34:19 UTC - in response to Message 1930480.  

Thanks Brent for this advice.
Cheers
ID: 1930483 · Report as offensive
Profile RueiKe Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 492
Credit: 378,512,430
RAC: 785
Taiwan
Message 1931019 - Posted: 20 Apr 2018, 10:47:48 UTC

I have been having trouble with the Linux r3584 SoG app hanging and the process never ending. Could be a system issue, but I wanted to try the ATi version of the r3584 Linux app before I make major changes to my system. Does it sound like a possibility that the SoG app is a contributing factor? Can someone point me to where I can download the app? Thanks!
GitHub: Ricks-Lab
Instagram: ricks_labs
ID: 1931019 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1931032 - Posted: 20 Apr 2018, 12:52:25 UTC - in response to Message 1931019.  

On my older ATI GPUs r3584 always gave False overflows. The 3505 version doesn't give False overflows and is just as fast as r3584. You might try that, ATI5r3505&IntelSSSE3CPU.7z Otherwise, you might try a different GPU driver. I've been in a battle with the last version of the Special App that works in OSX Yosemite and False overflows on a GTX 750 Ti. It seems a slightly older CUDA driver may have fixed it. Strange it only happens with the 750 Ti, the GTX 9 series doesn't have the problem.
ID: 1931032 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34363
Credit: 79,922,639
RAC: 80
Germany
Message 1931085 - Posted: 20 Apr 2018, 20:41:32 UTC
Last modified: 20 Apr 2018, 20:43:58 UTC

I`m running 3584 on Linux Mint 18.3 R9 380 and never had an issue nor invalid.


With each crime and every kindness we birth our future.
ID: 1931085 · Report as offensive
Profile RueiKe Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 492
Credit: 378,512,430
RAC: 785
Taiwan
Message 1931327 - Posted: 22 Apr 2018, 3:20:19 UTC - in response to Message 1931032.  

On my older ATI GPUs r3584 always gave False overflows. The 3505 version doesn't give False overflows and is just as fast as r3584. You might try that, ATI5r3505&IntelSSSE3CPU.7z Otherwise, you might try a different GPU driver. I've been in a battle with the last version of the Special App that works in OSX Yosemite and False overflows on a GTX 750 Ti. It seems a slightly older CUDA driver may have fixed it. Strange it only happens with the 750 Ti, the GTX 9 series doesn't have the problem.


Thanks TBar! I have downloaded, but this weekend I convinced my self that I have spent way too much time trying to get my Triple ProDuo block of GPU's working on my latest platform. This configuration has had issues with Zenith Extreme and X99-E-10G, but worked with no issues on the Crosshair V Formula-Z, which was its original home. I moved it back over the weekend and it has no problems. I have moved the block of 4 RX Vega64 over to my Linux cruncher only to realize that ROCm doesn't yet support X399 platform...

During the troubleshooting of the issues with this configuration, I went back and forth between 16.04.3 and 16.04.4 and AMD drivers from last quarter and the latest. I originally built it on the ASRock X399 professional with no GPU issues but had OC stability issues with it, which is why I moved it to Zenith.
GitHub: Ricks-Lab
Instagram: ricks_labs
ID: 1931327 · Report as offensive
Profile RueiKe Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 492
Credit: 378,512,430
RAC: 785
Taiwan
Message 1931328 - Posted: 22 Apr 2018, 3:26:46 UTC - in response to Message 1931085.  

I`m running 3584 on Linux Mint 18.3 R9 380 and never had an issue nor invalid.


Hi Mike, Thanks for sharing your experience for confirmation. I strongly suspect it is just the Triple ProDuo configuration that is the root of the problem. Confirmation on IG that another user with 2 ProDuo and 2 Vega64 has issues mining with all 6 GPUs on Zenith (Windows). Problem goes away by disabling one card. Getting this setup working on Zenith is a bit cludgy as it will get NVRAM error unless you bring it up with 1 card first from cleared CMOS. ASRock had no GPU issues, but just could not OC as well. Maybe Asus will have a BIOS fix, but I have been waiting 1 year for a confirmed issue I reported on C6H to be fixed, so I am not going to wait around...
GitHub: Ricks-Lab
Instagram: ricks_labs
ID: 1931328 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1947230 - Posted: 1 Aug 2018, 12:15:25 UTC

Since the Old thread is LOCKED.... I started a New one, https://setiathome.berkeley.edu/forum_thread.php?id=80636&postid=1931328

So, setiathome_x41p_V0.96_x86_64-Linux_cuda9x has made it's appearance and for the first time since zi3v, it works with the Overflows.
Right now it still has trouble with the Arecibo shorties, but, at present that's not a real problem. Currently Speed improvements are in the 30 to 60% range, depending on the GPU, with the 750 Ti being around 30 % faster. Stay tuned...

BTW, if someone wants to Unlock the old thread and move this post there it would be appreciated.
ID: 1947230 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1947249 - Posted: 1 Aug 2018, 13:46:40 UTC - in response to Message 1947230.  

i was trying v0.95 on 750ti's and it was just immediately trashing every work unit with computation errors.

v0.96 fixed this? where's the download link? your link just redirects back to your own post lol.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1947249 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1947265 - Posted: 1 Aug 2018, 14:37:57 UTC - in response to Message 1947249.  

Hmmmm, you must have a compile from Petri, I never compiled 0.95 on Linux. After getting Hundreds of Invalids on My Macs with 0.95 I decided it wasn't worth the trouble to find my old Linux dev system. I never did get Errors with 0.95, just Every Overflow ended up invalid... except a handful that cross validated with W3Perl. I never did find my old system either, had to build another Linux dev system. I have two machines running 0.96 on Linux with 750 Ti, both work fine. I'm still trying to figure out why they hate Arecibo shorties, once that is fixed I will post a link to the fully working App.
ID: 1947265 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1947276 - Posted: 1 Aug 2018, 15:07:06 UTC - in response to Message 1947265.  

Sounds good TBar. Will look for the app download link when it is ready for primetime.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1947276 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1947285 - Posted: 1 Aug 2018, 15:45:13 UTC - in response to Message 1947265.  
Last modified: 1 Aug 2018, 15:48:00 UTC

Hmmmm, you must have a compile from Petri, I never compiled 0.95 on Linux. After getting Hundreds of Invalids on My Macs with 0.95 I decided it wasn't worth the trouble to find my old Linux dev system. I never did get Errors with 0.95, just Every Overflow ended up invalid... except a handful that cross validated with W3Perl. I never did find my old system either, had to build another Linux dev system. I have two machines running 0.96 on Linux with 750 Ti, both work fine. I'm still trying to figure out why they hate Arecibo shorties, once that is fixed I will post a link to the fully working App.


is there a link for v0.96 somewhere, or should i ask petri for it? like i said, your link in your previous posts just redirects back to itself. ..

v0.95 runs fine on my 2* 1050ti machine. with a 38% speed boost vs the "normal" special app. i do see issues with invalid short tasks. no errors, just end up invalid. all on the order of <30s run times. the longer ones process just fine and validate.

v0.95 just doesnt work on 750ti's. it's not much consequence, im upgrading that sytem to dual 1060's soon enough.

and for anyone else not aware, v0.95+ requires R396+ beta nv drivers. 390 whql wont work.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1947285 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1947319 - Posted: 1 Aug 2018, 21:18:23 UTC

Hi,

the 0.96 is still not finished. Like TBar said there are problems with Arecibo shorties. When we get them ironed out we will come up with V1.00 .

I've got two weeks summer vacation left and I hope I have time to fix the app.

Petri
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1947319 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1947425 - Posted: 2 Aug 2018, 4:34:42 UTC

that would be awesome. we really appreciate the effort to improve our performance!
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1947425 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1947722 - Posted: 3 Aug 2018, 16:14:42 UTC

Looking forward to the newest app. Was browsing some of Petri's tasks and came upon this one.
https://setiathome.berkeley.edu/workunit.php?wuid=3079506159

which just so happened to be run on both hosts by a 1080 card. Petri's host finished it 60 seconds faster than an older x41p_zi3t2b application. Granted an older CUDA8 app vs. CUDA9.2 app but the statically linked app is so much faster it will be nice to use eventually.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1947722 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1947741 - Posted: 3 Aug 2018, 19:09:59 UTC

@Petri, can you explain simply what the new tuning parameter in the latest 0.96 Beta does. I am referring to:

{Using default pulse Fft limit (-pfl 64)} in the stderr.txt output for a task. This seems to be a new parameter not mentioned in the original x41zi or x41p_zi3v notes.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1947741 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1947748 - Posted: 3 Aug 2018, 20:01:23 UTC - in response to Message 1947741.  

Petri did mention the -pfl flag before in https://setiathome.berkeley.edu/forum_thread.php?id=78569&postid=1944686
<snip>
Could you try with a new flag: -pfl 64
It sets a splitting point to pulse finding which one of the internal functions to use.
I find the -pfl 64 give faster run times compared to default (no flag set equals to -pfl 512). I'll change that in confsettings.cpp to 64 in the future.
ID: 1947748 · Report as offensive
Previous · 1 . . . 79 · 80 · 81 · 82 · 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.