V8 CUDA for Linux?

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

To post messages, you must log in.

Previous · 1 . . . 7 · 8 · 9 · 10

AuthorMessage
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1783779 - Posted: 30 Apr 2016, 15:18:55 UTC - in response to Message 1783771.  

...So seems to me we have some technological convergence going on between Windows and Linux :)

So Jason, in your opinion, is this a good thing or a bad thing? Are they 'dumbing down' (for lack of a better way of describing it) Linux which might reduce it's performance and implied (to me) advantages over Windows? Or is it good in that it allows programmers to work with more of a single skill set? Or something completely different?


It's hard (speaking as a computer scientist). The *NIX model as has been has some 30+years behind it (many more if you count BCPL, predecessor to C etc). At the same time, I believe *NIX dropped the ball with threading and scalability, which is becoming a thing finally.

These are things that M$ did before *NIX, but then the commodore Amiga had preemptive multitasking before either of them anyway (despite that the Amiga was based on a SunStation)... So It comes down to precious developers protecting their turf and being precious.

In this case I think the CPU scaling stall was ineveitable, and the only choice was multithreaded everything. For what we have now it's just M$ that recognised it first, and embraced it early (~2005);
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1783779 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1784088 - Posted: 1 May 2016, 20:57:52 UTC
Last modified: 1 May 2016, 21:16:34 UTC

Anyone have any ideas?

Anyone?

Well after fooling around a bit longer with a different configure line it seems to be the libwx version used is just not working to build the manager. That's after you get by all the "can't find file blah blah blah" junk.
Funny, when I look at BOINC 7.4.25 I don't see any libwx dependencies. So, it must be possible to build a BOINC without all the libwx BS and 7.4.25 does work in Ubuntu 16.04...mostly. I suppose you would need the files from around late 2014?

Right now as far as you get is;
  CXXLD    boincmgr
/usr/bin/ld: boincmgr-taskbarex.o: undefined reference to symbol 'gtk_status_icon_get_icon_name'
/usr/lib/x86_64-linux-gnu/libgtk-x11-2.0.so.0: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status
Makefile:686: recipe for target 'boincmgr' failed
....

Once you nuke that error there are a few similar ones waiting.
I still haven't figured out why I need a taskbar ex in Ubuntu...
ID: 1784088 · Report as offensive
ShootY
Avatar

Send message
Joined: 15 Jan 11
Posts: 71
Credit: 21,700,959
RAC: 41
Germany
Message 1784415 - Posted: 3 May 2016, 5:47:03 UTC
Last modified: 3 May 2016, 5:49:45 UTC

Crunching on nvidia Cuda isn't supported yet (officially).
Eric Korpela once postet a link where you can find the latest "app version" of Seti supported hardware => http://setiathome.berkeley.edu/apps.php
There's no nvidia Cuda support on Linux yet.

But even on AMD/ATi isn't running on my PC (Linux).

But in his post from April 12, he wrote:

The VLARs go out to ATI GPUs, but are held back from NVIDIA. If I could find a way to send VLAR to only NVIDIA OpenCL and not NVIDIA CUDA, I would...
(Link: https://setiathome.berkeley.edu/forum_thread.php?id=79411&postid=1778449#1778449)

"No usable GPUs found"

No idea why...
ID: 1784415 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1784437 - Posted: 3 May 2016, 6:37:06 UTC - in response to Message 1784415.  
Last modified: 3 May 2016, 6:59:23 UTC

Actually, AMD cards have been working in Linux since Jan 22. All this headbanging trying to build a recent version of BOINC on Linux seems to have rattled my brain.
I was running it a couple weeks earlier than Jan 22 with the same App I'm using now, http://setiathome.berkeley.edu/results.php?hostid=7769537&offset=160
For me all it takes is that one link in /usr/lib.

The problem on the AMD cards is the link to the OpenCL library. You might try a solution posted earlier that takes the shotgun approach and lists a few different directories in path; https://setiathome.berkeley.edu/forum_thread.php?id=78279&postid=1740846#1740846
Adding LD_LIBRARY_PATH=/lib64:/usr/lib:/lib:/usr/local/lib to etc/default/ did the trick.

One of those directories is where BOINC should be looking for the OpenCL library.

There isn't a problem running MB tasks on your Linux nVidia card either. There are Apps available and simply installing the recent driver from nVidia allows the card to work in Linux.
ID: 1784437 · Report as offensive
ShootY
Avatar

Send message
Joined: 15 Jan 11
Posts: 71
Credit: 21,700,959
RAC: 41
Germany
Message 1784449 - Posted: 3 May 2016, 7:25:13 UTC

Ok will test this: Adding LD_LIBRARY_PATH=/lib64:/usr/lib:/lib:/usr/local/lib to etc/default/

What driver would you recommend?
- Proprietary driver (fglrx/fglrx-update)
or
- xorg-driver (open source)
ID: 1784449 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1784458 - Posted: 3 May 2016, 8:03:29 UTC - in response to Message 1784449.  

You have to use a driver with OpenCL, the x-org driver does Not have OpenCL. The fglrx ones do. I'm using an older driver from AMD as it seems to work best on my older cards.

I just looked up that "fix". Apparently it's missing the last part; http://boinc.berkeley.edu/dev/forum_thread.php?id=10315&postid=62392#62392
LD_LIBRARY_PATH=/lib64:/usr/lib:/lib:/usr/local/lib
to /etc/defaults/boinc-client which started GPU tasks running.

It seemed something was missing. Note, this fix appears to be for people running the BOINC App from the Repository, Not the Berkeley downloaded version of BOINC which I'm using.
ID: 1784458 · Report as offensive
ShootY
Avatar

Send message
Joined: 15 Jan 11
Posts: 71
Credit: 21,700,959
RAC: 41
Germany
Message 1784460 - Posted: 3 May 2016, 8:45:38 UTC - in response to Message 1784458.  
Last modified: 3 May 2016, 8:46:05 UTC

Ok I'll try it when I'm home.
Last question: Do I have to type the command just in the shell or is it necessary beeing in the folder lib or so? Sorry, I'm new to linux... ;)
Would you please be so kind to post exactly what I have to type in?
Thank you!
ID: 1784460 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1784467 - Posted: 3 May 2016, 9:39:01 UTC - in response to Message 1784437.  
Last modified: 3 May 2016, 9:39:55 UTC

One of those directories is where BOINC should be looking for the OpenCL library.

Or update to BOINC 7.6.31 which looks for both libOpenCL.so to libOpenCL.so.1 libraries on its own.
Locutus of Borg's PPA has 7.6.31 for different distros, while some Ubuntu forms have it from repository.
ID: 1784467 · Report as offensive
ShootY
Avatar

Send message
Joined: 15 Jan 11
Posts: 71
Credit: 21,700,959
RAC: 41
Germany
Message 1784561 - Posted: 3 May 2016, 20:56:09 UTC - in response to Message 1784458.  


I just looked up that "fix". Apparently it's missing the last part; http://boinc.berkeley.edu/dev/forum_thread.php?id=10315&postid=62392#62392
LD_LIBRARY_PATH=/lib64:/usr/lib:/lib:/usr/local/lib
to /etc/defaults/boinc-client which started GPU tasks running.

It seemed something was missing. Note, this fix appears to be for people running the BOINC App from the Repository, Not the Berkeley downloaded version of BOINC which I'm using.



I have the version from the Repository but when I type in this codes it still doesn't work to me.
Maybe I did something wrong - as I said I'm new to Linux.
Could you please give me a step by step solution what to type into the shell?
Thank you
ID: 1784561 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1784566 - Posted: 3 May 2016, 21:15:59 UTC - in response to Message 1784561.  

Well, as I said, I don't use the Repository version of BOINC. However, /etc/default is a directory and boinc-client should be another file similar to the others there. I would open boinc-client with a text editor and paste in LD_LIBRARY_PATH=/lib64:/usr/lib:/lib:/usr/local/lib below any other variables already there. Then restart the machine.
ID: 1784566 · Report as offensive
ShootY
Avatar

Send message
Joined: 15 Jan 11
Posts: 71
Credit: 21,700,959
RAC: 41
Germany
Message 1784652 - Posted: 4 May 2016, 5:20:43 UTC - in response to Message 1784566.  

Ok did it.
Still no usable GPUs found. The OpenCL (AMD) is recognized - but not the GPU itself. Weird. But thank you so much for helping me! I really appreciate that!
ID: 1784652 · Report as offensive
ShootY
Avatar

Send message
Joined: 15 Jan 11
Posts: 71
Credit: 21,700,959
RAC: 41
Germany
Message 1784744 - Posted: 4 May 2016, 17:00:43 UTC
Last modified: 4 May 2016, 17:11:43 UTC

Found a solution - but still not working as expected:

Solution found here:
https://boinc.berkeley.edu/dev/forum_thread.php?id=6307#36496



This worked for me:


BOINC starts before GDM can finish getting up and running. This makes the BOINC think the video card is absent. This can occur with both ATI and Nvidia cards but I first found it on Nvidia card equipped machines. If your GPU is recognized after you restart BOINC but doesn't find the card right after a reboot this is likely your problem. This can be a bit sporadic on some machines where it'll find it after one reboot but not another. If this is the only problem it'll find it every time after a restart of just BOINC.

sudo /etc/init.d/boinc-client restart

See history here: https://bugs.launchpad.net/ubuntu/+source/boinc/+bug/414244



BUT

FIX #1 did NOT work to me!

FIX #1: My fix was to add a delay into the start-up script so that GDM/X has a chance to have drivers fully loaded. This solution inserts a 6 second delay in the boinc start-up script which allows ample time for the driver to be available on all machines tested to date.
(a)Edit the start-up script file /etc/init.d/boinc-client with sudo gedit /etc/init.d/boinc-client.
(b)Find this function code and add the line “sleep 6” where it's shown here:
start()
{
log_begin_msg "Starting $DESC: $NAME"
if is_running; then
log_progress_msg "already running"
else
sleep 6
start-stop-daemon --start --quiet --background --pidfile $PIDFILE \
--make-pidfile --user $BOINC_USER --chuid $BOINC_USER \
--chdir $BOINC_DIR --exec $BOINC_CLIENT -- $BOINC_OPTS
fi
log_end_msg 0

if [ "$SCHEDULE" = "1" ]; then
schedule
fi
}


FIX #2 didn't work either.


This means I always have to restart Boinc-Client.

ID: 1784744 · Report as offensive
Profile tazzduke
Volunteer tester

Send message
Joined: 15 Sep 07
Posts: 190
Credit: 28,269,068
RAC: 5
Australia
Message 1785160 - Posted: 6 May 2016, 8:26:00 UTC

Good Afternoon

I am now running a 750ti on one my computers, I hope to have two more into my other computers within 2 weeks depending on post.

Question is it viable to run 2 at a time on these cards, as I have added the poll command that was discussed in another thread for NVIDIA cards and I am getting around 90% usage.

Probably wrong thread, but it was another way of saying hello as well.....

The GTX 960 and GTX 670 have been sold off as I was looking for a more economical solution to continue crunching full time.

Okay, to those who provide some input, I thankyou in advance

Regards
Mark
ID: 1785160 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1785164 - Posted: 6 May 2016, 9:31:33 UTC - in response to Message 1785160.  

You can try running more instances, but, once you get above 90% GPU usage there isn't much advantage. Each new instance will use another full CPU core with very little advantage, you will basically be doubling the run time and wasting a CPU core that could be running a CPU task. I was getting a little better GPU usage with the CUDA 42 App and the run times were better as a result, http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=72013&offset=40
ID: 1785164 · Report as offensive
Profile tazzduke
Volunteer tester

Send message
Joined: 15 Sep 07
Posts: 190
Credit: 28,269,068
RAC: 5
Australia
Message 1785166 - Posted: 6 May 2016, 9:41:57 UTC

Greetings TBar

Thankyou for the reply, I will give the CUDA42 a shot over the weekend and then go from there.

Gonna stick with 1 instance then as well.

Regards
ID: 1785166 · Report as offensive
OTS
Volunteer tester

Send message
Joined: 6 Jan 08
Posts: 369
Credit: 20,533,537
RAC: 0
United States
Message 1785266 - Posted: 6 May 2016, 18:35:22 UTC - in response to Message 1785164.  

You can try running more instances, but, once you get above 90% GPU usage there isn't much advantage. Each new instance will use another full CPU core with very little advantage, you will basically be doubling the run time and wasting a CPU core that could be running a CPU task. I was getting a little better GPU usage with the CUDA 42 App and the run times were better as a result, http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=72013&offset=40



FWIW, I run two V8 MBs at a time on my 750ti and the card runs at 95% utilization plus or minus with a temp of around 48C. I only keep one CPU core free for both WU’s and that core shows about 15% utilization on average. This is with app setiathome_x41zi_x86_64-pc-linux-gnu_cuda60. I am not sure I gain a whole lot though as the time for each WU nearly doubles when you run two tasks at a time. At least that is what I recall. I have been running two WUs for many months now and I would have to run the one versus two WU tests again to be sure.
ID: 1785266 · Report as offensive
Profile Francesco Forti
Avatar

Send message
Joined: 24 May 00
Posts: 334
Credit: 204,421,005
RAC: 15
Switzerland
Message 1785787 - Posted: 8 May 2016, 8:28:54 UTC - in response to Message 1783434.  
Last modified: 8 May 2016, 8:29:04 UTC

It just appeared for me last night (Mint 17.3); packages are cuda-drivers, nvidia-graphics-drivers-352, nvidia-modprobe and nvidia-settings in the Mint Update Manager all 352.93.

I confirm.
Now I see nvidia-modprobe too. Was missing some weeks ago.
ID: 1785787 · Report as offensive
Previous · 1 . . . 7 · 8 · 9 · 10

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.