Setting up Linux to crunch CUDA90 and above for Windows users

Message boards : Number crunching : Setting up Linux to crunch CUDA90 and above for Windows users
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 . . . 162 · Next

AuthorMessage
Profile scocam
Avatar

Send message
Joined: 28 Feb 17
Posts: 27
Credit: 15,120,999
RAC: 0
United States
Message 1861299 - Posted: 13 Apr 2017, 21:29:19 UTC

Many thanks to Petri and TBar for sharing their hard work and knowledge and to Stephen for the instruction.
I'm looking forward to building my new linux cruncher this weekend (details below). With some luck, I hope to be up and running by Monday. Thank again, all.

Regards,
scocam


New build:
Intel i7 6850k
ASUS X99-E WS
EVGA GTX 1070 (3 or 4)
Corsair Dominator Platinum 32gb
EVGA SuperNOVA 1600 G2
Water cooled GPUs/CPU
ID: 1861299 · Report as offensive     Reply Quote
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 1861322 - Posted: 13 Apr 2017, 23:59:11 UTC - in response to Message 1861299.  

Many thanks to Petri and TBar for sharing their hard work and knowledge and to Stephen for the instruction.
I'm looking forward to building my new linux cruncher this weekend (details below). With some luck, I hope to be up and running by Monday. Thank again, all.

Regards,
scocam


New build:
Intel i7 6850k
ASUS X99-E WS
EVGA GTX 1070 (3 or 4)
Corsair Dominator Platinum 32gb
EVGA SuperNOVA 1600 G2
Water cooled GPUs/CPU


. . Enjoy and have a smooth and speedy build :)

Stephen

PS: the number shown as 3 or 4 against the GPU's intrigues me. Did you lose count ? :)

:)
ID: 1861322 · Report as offensive     Reply Quote
Profile scocam
Avatar

Send message
Joined: 28 Feb 17
Posts: 27
Credit: 15,120,999
RAC: 0
United States
Message 1861325 - Posted: 14 Apr 2017, 0:31:36 UTC - in response to Message 1861322.  

If only I had so many gpus to lose count! I'm not sure how many gpus I'm going to rob from my other machine quite yet. I have two 1070s waiting on this build but I may borrow a 1 or 2 more from my current machine. If crunching with cuda80 is as efficient as it looks, I'll load this new build with as many gpus as I can.

Regards,
scocam
ID: 1861325 · Report as offensive     Reply Quote
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 1861330 - Posted: 14 Apr 2017, 0:57:24 UTC - in response to Message 1861325.  
Last modified: 14 Apr 2017, 1:05:48 UTC

If only I had so many gpus to lose count! I'm not sure how many gpus I'm going to rob from my other machine quite yet. I have two 1070s waiting on this build but I may borrow a 1 or 2 more from my current machine. If crunching with cuda80 is as efficient as it looks, I'll load this new build with as many gpus as I can.

Regards,
scocam


. . OOoooooohhh it is! The only shortcoming I am aware of, is when confronted with lots of shorties it can tend to output a high ratio of inconclusives (about 7 to 10%) due to an as yet unresolved ecentricity in the order of outputting results. But for the most part (about 99.9%) they validate AOK. So overall the productivity justifies the app.

. . Pentium-D rig with 2 x 1060-6GB cards, under SoG between 32 and 39K, with CUDA80 and as yet not settled about 52K.
. . Core2 Duo with one low profile standard clocked 1050ti, under SoG 13K, so far with CUDA80 22K.

. . Now we all know that RAC is not consistent, and not an imperial measure of productivitty because of that, but it remains the most common available means of comparison we have. And with all it's shortcomings it speaks volumes here :)

. . And while at this point I may be a tad optimistic, if/when it gets approved for general release, we may see DA's concerns, about insufficient processing power available to the project in the broader BOINC/Seti community, evaporate. With the burgeoning population of GTX1070 and GTX1080 cards combined with this app we could see an exponential growth in crunching capacity around the globe. Heck even with humble GTX1050ti's popping up like little mushrooms we could see a massive growth.

. . Enjoy!

Stephen

:)
ID: 1861330 · Report as offensive     Reply Quote
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 1861357 - Posted: 14 Apr 2017, 3:40:17 UTC - in response to Message 1861325.  

My 1070 was running around 73s for shorties - you definitely won't see that with SoG.
ID: 1861357 · Report as offensive     Reply Quote
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 1861361 - Posted: 14 Apr 2017, 3:54:22 UTC - in response to Message 1861357.  

My 1070 was running around 73s for shorties - you definitely won't see that with SoG.


. . With my 1060s shorties, or halflings as I like to call them to differentiate them from noise bombs (which run for only a few seconds), take about 100 secs, compared to full length normal AR tasks which take about 210 to 220 seconds (hence halflings). Pretty good as far as I am concerned.

Stephen

:)
ID: 1861361 · Report as offensive     Reply Quote
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 1861364 - Posted: 14 Apr 2017, 4:06:34 UTC - in response to Message 1861361.  

Thinking, that was actually detuned because of a card mismatch - 980 w/15 CU, 1070 w/16 CU.
Running with unroll 15 for the 980, the 1070 was only at 94% capacity.
ID: 1861364 · Report as offensive     Reply Quote
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 1861365 - Posted: 14 Apr 2017, 4:06:39 UTC
Last modified: 14 Apr 2017, 4:08:53 UTC

. . An observation on using the SETI based version of BOINC and installing to /home/BOINC as opposed to using the repository version which installs to /var/lib/boinc-client and /etc/boinc-client. The rep version does nost use the identity root as I had thought, but rather creates an identity boinc for ownership of the crunching apps. But the advantage with this version is that it not only creates the daemon to launch boinc client but also creates the icon for boinc manager on the launch bar allowing you to lock it there and use it to launch manager whenever you require it. With the "preferred" version which makes your logon identity the owner, while it does create the daemon to launch the client it does not create the icon for manager on the launch bar. On my system this only appears once you launch boinc manager from a terminal window and does not allow you to lock it there. I am trying to find out how to create the init file to make it possible to do that, or modify the existing files if this can be done.

Stephen

??
ID: 1861365 · Report as offensive     Reply Quote
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 1861370 - Posted: 14 Apr 2017, 4:16:15 UTC - in response to Message 1861365.  

I'm going to try the BOINC version of BOINC tonight (new SSD showed up) rather than the Ubuntu version and see how I like it.
I would imagine it should be able to lock to the launcher if it is started from the search app thingy (LOL I forget its name).
ID: 1861370 · Report as offensive     Reply Quote
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 1861371 - Posted: 14 Apr 2017, 4:18:38 UTC - in response to Message 1861364.  

Thinking, that was actually detuned because of a card mismatch - 980 w/15 CU, 1070 w/16 CU.
Running with unroll 15 for the 980, the 1070 was only at 94% capacity.


. . Using 'nvidia-smi -l' to display GPU usage on my 1060-6GB cards I see highly varying GPU usage using unroll of 10 which suits both cards which have CU=10. It varies between 75% and 100%, so I am not sure that 94% is actually a problem, but then I am running on a Noah epoch PC :)

. . On that subject I am looking forward to saving my pennies and updating to Ryzen-1700 when the dust has settled.

Stephen

:)
ID: 1861371 · Report as offensive     Reply Quote
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 1861375 - Posted: 14 Apr 2017, 4:45:08 UTC - in response to Message 1861370.  

I'm going to try the BOINC version of BOINC tonight (new SSD showed up) rather than the Ubuntu version and see how I like it.
I would imagine it should be able to lock to the launcher if it is started from the search app thingy (LOL I forget its name).


. . I don't think I ever knew its name, but I hadn't thought of that, I wonder if that will make a difference? Hang five ...

. . OK, it took a while but I launched it from that thingy and you are 100% correct. I could then lock it to the launch bar, problem solved. Well done. Thanks.

Stephen

:)
ID: 1861375 · Report as offensive     Reply Quote
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1861390 - Posted: 14 Apr 2017, 6:15:38 UTC - in response to Message 1861364.  

Thinking, that was actually detuned because of a card mismatch - 980 w/15 CU, 1070 w/16 CU.
Running with unroll 15 for the 980, the 1070 was only at 94% capacity.


-unroll autotune helps systems with heterogenous GPUs.

... and the high ar tasks take about 33 seconds on 1080Ti.
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: 1861390 · Report as offensive     Reply Quote
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 1861392 - Posted: 14 Apr 2017, 6:47:00 UTC - in response to Message 1861390.  

33 sec? Let me know next week what the Titan XP does :)

Yes I know the autotune thing, but was using zi3k+ which doesn't have it. I just installed your x41p_zi3t1f with autotune.

Question are you running with or without -bs ? (sorry, laughing moment) You didn't mention -bs in the command line when you posted the app.
ID: 1861392 · Report as offensive     Reply Quote
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1861407 - Posted: 14 Apr 2017, 9:00:36 UTC - in response to Message 1861392.  
Last modified: 14 Apr 2017, 9:04:01 UTC

33 sec? Let me know next week what the Titan XP does :)

Yes I know the autotune thing, but was using zi3k+ which doesn't have it. I just installed your x41p_zi3t1f with autotune.

Question are you running with or without -bs ? (sorry, laughing moment) You didn't mention -bs in the command line when you posted the app.


No bs :)
I have 12 cores on CPU and use only 6 of them for CPU tasks. -bs can help systems with less cores or systems with heat issues.
The option -bs has its name carefully thought. It is a counter strike to name SoG.
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: 1861407 · Report as offensive     Reply Quote
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1861418 - Posted: 14 Apr 2017, 11:13:25 UTC
Last modified: 14 Apr 2017, 11:15:14 UTC

Hi,
for anyone running my cuda app there is a new version to try. Especially if you are experiencing problems with BLC or Arecibo low ar WUs. A big thank you to Gianfranco for testing.

Please backup your current working system and test off line first.

The executable is version zi3t2b and it can be run on sm_35, 50, 52, and 61. (750,780,980,1080 and likes).
With 1 Mb of GPU ram you need -unroll 1. Other can use -unroll autotune.
Use -bs to reduce CPU usage.
Set -pfb to 8, 16 or 32.

You can get your copy of zi3t2b executable here and the source code is here

Please PM any success or failure.

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: 1861418 · Report as offensive     Reply Quote
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 1861424 - Posted: 14 Apr 2017, 12:00:51 UTC - in response to Message 1861407.  
Last modified: 14 Apr 2017, 12:08:38 UTC

33 sec? Let me know next week what the Titan XP does :)

Yes I know the autotune thing, but was using zi3k+ which doesn't have it. I just installed your x41p_zi3t1f with autotune.

Question are you running with or without -bs ? (sorry, laughing moment) You didn't mention -bs in the command line when you posted the app.


No bs :)
I have 12 cores on CPU and use only 6 of them for CPU tasks. -bs can help systems with less cores or systems with heat issues.
The option -bs has its name carefully thought. It is a counter strike to name SoG.


. . Does that mean I should look into -bs for my rig with the 2 core Pentium CPU? FWIIW, it is running on defaults at the moment and is using between 65% and 80% of CPU (mainly around 70/75). Which is fine because there is nothing else for it to do :) And the CPU temps vary between 32 and 38, so no heat problem.

Stephen

?
ID: 1861424 · Report as offensive     Reply Quote
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1861426 - Posted: 14 Apr 2017, 12:07:49 UTC - in response to Message 1861424.  

33 sec? Let me know next week what the Titan XP does :)

Yes I know the autotune thing, but was using zi3k+ which doesn't have it. I just installed your x41p_zi3t1f with autotune.

Question are you running with or without -bs ? (sorry, laughing moment) You didn't mention -bs in the command line when you posted the app.


No bs :)
I have 12 cores on CPU and use only 6 of them for CPU tasks. -bs can help systems with less cores or systems with heat issues.
The option -bs has its name carefully thought. It is a counter strike to name SoG.


. . Does that mean I should look into -bs for my rig with the 2 core Pentium CPU?

Stephen

?


Only testing can tell. Set the -bs flag and watch what happens to run times.
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: 1861426 · Report as offensive     Reply Quote
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 1861428 - Posted: 14 Apr 2017, 12:12:44 UTC - in response to Message 1861426.  
Last modified: 14 Apr 2017, 12:20:35 UTC



. . Does that mean I should look into -bs for my rig with the 2 core Pentium CPU?

?


Only testing can tell. Set the -bs flag and watch what happens to run times.


. . OK, where do I set that? I have not looked into settings other than defaults. I have looked at app_info.xml and found the line ...

<cmdline> -bs -unroll 10 </cmdline>

. . So I take it that it is already set. Do you want me to take it out and see what difference it makes? Or just leave well enough alone :)

Stephen

?
ID: 1861428 · Report as offensive     Reply Quote
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1861431 - Posted: 14 Apr 2017, 12:23:49 UTC - in response to Message 1861428.  



. . Does that mean I should look into -bs for my rig with the 2 core Pentium CPU?

?


Only testing can tell. Set the -bs flag and watch what happens to run times.


. . OK, where do I set that? I have not looked into settings other than defaults. I have looked at app_info.xml and found the line ...

<cmdline> -bs -unroll 10 </cmdline>

. . So I take it that it is already set. Do you want me to take it out and see what difference it makes? Or just leave well enough alone :)

Stephen

?


I'm sure a lot of people would like to know if and how it effects the performance.
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: 1861431 · Report as offensive     Reply Quote
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 1861501 - Posted: 14 Apr 2017, 22:30:59 UTC - in response to Message 1861431.  



Only testing can tell. Set the -bs flag and watch what happens to run times.


. . OK, I have looked at app_info.xml and found the line ...

<cmdline> -bs -unroll 10 </cmdline>

. . So I take it that it is already set. Do you want me to take it out and see what difference it makes? Or just leave well enough alone :)

Stephen

?


I'm sure a lot of people would like to know if and how it effects the performance.


. . OK I will remove the -bs flag and check the outcomes. Which rig would you like me to try, the one with one 1050ti or the one with the two 1060s?

Stephen

?
ID: 1861501 · Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 . . . 162 · Next

Message boards : Number crunching : Setting up Linux to crunch CUDA90 and above for Windows users


 
©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.