Deprecated: Function get_magic_quotes_gpc() is deprecated in /disks/centurion/b/carolyn/b/home/boincadm/projects/beta/html/inc/util.inc on line 663
Need bullshit guide for GPU computing on Linux!

Need bullshit guide for GPU computing on Linux!

Message boards : SETI@home Enhanced : Need bullshit guide for GPU computing on Linux!
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Tetsuji Maverick Rai
Project developer
Volunteer developer
Avatar

Send message
Joined: 15 Jun 05
Posts: 399
Credit: 16,571,350
RAC: 0
Japan
Message 54422 - Posted: 13 Jun 2015, 11:37:22 UTC
Last modified: 13 Jun 2015, 11:58:01 UTC

I cannot believe I have this question!

Somehow I cannot run GPU applications on Linux running on i5 w/ nVidia GeForce 750ti (On Windows 8.1/7, they run as expected). I have installed Fedora 22 and did "dnf update" to refresh packages, disabled nouveau, and installed nVidia driver, so called NVIDIA-Linux-x86_64-346.72.run and edit /etc/X11/xorg.conf to insert Load "glx". Actually glxinfo says it can do direct rendering, so nvidia driver is loaded. If I set GPU to "always run" it gives computation error, but if I set GPU to "To preference", it never starts.

boinc-client and boinc-manager rpm packages are also loaded and set boinc by systemctl enable blah blah..... So boinc manager says many wu's are downloaded, including for GPU, but wu's for GPU never start.

What's wrong with my config? These 10 years I have never programmed for seti@home.... :(

Thank you very much.
Luckiest in the world. WMD = Weapon of Mass Distraction
ID: 54422 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 2 Jul 13
Posts: 505
Credit: 5,019,318
RAC: 0
United States
Message 54423 - Posted: 13 Jun 2015, 14:35:54 UTC - in response to Message 54422.  

It appears you either have a couple of identical machines or identical hosts. Looking at this host the Error says;
../../projects/setiweb.ssl.berkeley.edu_beta/setiathome_7.07_x86_64-pc-linux-gnu__opencl_nvidia_sah: error while loading shared libraries: libOpenCL.so.1: cannot open shared object file: No such file or directory
Sounds as though the driver isn't working correctly even though BOINC says it has OpenCL.
What happens if you run clinfo in the terminal?
ID: 54423 · Report as offensive
Urs Echternacht
Volunteer tester
Avatar

Send message
Joined: 18 Jan 06
Posts: 1038
Credit: 18,734,730
RAC: 0
Germany
Message 54424 - Posted: 13 Jun 2015, 16:23:32 UTC - in response to Message 54422.  
Last modified: 13 Jun 2015, 16:23:55 UTC

I cannot believe I have this question!
...snip...
What's wrong with my config? These 10 years I have never programmed for seti@home.... :(

Thank you very much.

Set a symbolic link to your libOpenCL, however it is named on your distro, and name that link like the missing library named in your error message.


Why does every new version of a distro try to mix its own soup of lib-names ? That makes it hard for users, but impossible for an app-developer to anticipate.

In this case we have to care for RHEL6 users, which are quite a big group on Linux, that use link named "libOpenCL.so.1"
But it looks like Fedora22, which is a RHEL7 derived distro, does use links to libs differently named.
_\|/_
U r s
ID: 54424 · Report as offensive
Profile Tetsuji Maverick Rai
Project developer
Volunteer developer
Avatar

Send message
Joined: 15 Jun 05
Posts: 399
Credit: 16,571,350
RAC: 0
Japan
Message 54425 - Posted: 13 Jun 2015, 18:14:45 UTC

Thank you, TBar and Urs Echternacht for your suggestions. I don't have two identical machines :) Just forgot to merge.

BTW, libOpenCL.so.1 exists in /usr/lib. There is a shared library libOpenCL.so.1.0.0 and 3 symlinks to it, libOpenCL.so.1.0, libOpenCL.so.1, libOpenCL.so.

If I run ./seti*opencl*sah in standalone (w/o wu), it produces stderr.txt (which means it loads shared libraries), which says
[root@localhost setiweb.ssl.berkeley.edu_beta]# cat stderr.txt
03:06:02 (3352): Can't open init data file - running in standalone mode
03:06:02 (3352): Can't open init data file - running in standalone mode
Not using mb_cmdline.txt-file, using commandline options.
03:06:02 (3352): Can't open init data file - running in standalone mode
WARNING: init_data.xml missing
OpenCL platform detected: NVIDIA Corporation
WARNING: BOINC supplied wrong platform!
Number of OpenCL devices found : 1 
BOINC assigns slot on device #0.
WARNING: BOINC failed to provide OpenCL device, using own enumeration abilities
OpenCL queue synchronized
SETI@home error -5 Can't open file
(work_unit.sah) in read_wu_state() errno=2

File: worker.cpp
Line: 136


What's the problem?

Thanks in advance.
Luckiest in the world. WMD = Weapon of Mass Distraction
ID: 54425 · Report as offensive
Profile Tetsuji Maverick Rai
Project developer
Volunteer developer
Avatar

Send message
Joined: 15 Jun 05
Posts: 399
Credit: 16,571,350
RAC: 0
Japan
Message 54426 - Posted: 13 Jun 2015, 20:11:15 UTC
Last modified: 13 Jun 2015, 20:13:49 UTC

btw, I usually use runlevel 3 (w/o X) when log into Linux, and run startx to bring up X server. Does it matter? Of course nvidia module is always with the kernel.
Luckiest in the world. WMD = Weapon of Mass Distraction
ID: 54426 · Report as offensive
Profile Tetsuji Maverick Rai
Project developer
Volunteer developer
Avatar

Send message
Joined: 15 Jun 05
Posts: 399
Credit: 16,571,350
RAC: 0
Japan
Message 54427 - Posted: 13 Jun 2015, 20:24:08 UTC

I fresh installed Linux from scratch, and now "Use GPU always" didn't crash GPU based s@h(opencl_nvidia_sah). But I'm not sure when it runs. I will leave it as it is from now on.
Luckiest in the world. WMD = Weapon of Mass Distraction
ID: 54427 · Report as offensive
Urs Echternacht
Volunteer tester
Avatar

Send message
Joined: 18 Jan 06
Posts: 1038
Credit: 18,734,730
RAC: 0
Germany
Message 54429 - Posted: 13 Jun 2015, 22:26:54 UTC - in response to Message 54425.  
Last modified: 13 Jun 2015, 22:27:14 UTC

Thank you, TBar and Urs Echternacht for your suggestions. I don't have two identical machines :) Just forgot to merge.

BTW, libOpenCL.so.1 exists in /usr/lib. There is a shared library libOpenCL.so.1.0.0 and 3 symlinks to it, libOpenCL.so.1.0, libOpenCL.so.1, libOpenCL.so.

Check "/usr/lib64/" directory on 64bit linux.

If I run ./seti*opencl*sah in standalone (w/o wu), it produces stderr.txt (which means it loads shared libraries), which says
[root@localhost setiweb.ssl.berkeley.edu_beta]# cat stderr.txt
03:06:02 (3352): Can't open init data file - running in standalone mode
03:06:02 (3352): Can't open init data file - running in standalone mode
Not using mb_cmdline.txt-file, using commandline options.
03:06:02 (3352): Can't open init data file - running in standalone mode
WARNING: init_data.xml missing
OpenCL platform detected: NVIDIA Corporation
WARNING: BOINC supplied wrong platform!
Number of OpenCL devices found : 1 
BOINC assigns slot on device #0.
WARNING: BOINC failed to provide OpenCL device, using own enumeration abilities
OpenCL queue synchronized
SETI@home error -5 Can't open file
(work_unit.sah) in read_wu_state() errno=2

File: worker.cpp
Line: 136


What's the problem?

Thanks in advance.

No problem. That error -5 only describes that the "work_unit.sah" file is not there.
_\|/_
U r s
ID: 54429 · Report as offensive

Message boards : SETI@home Enhanced : Need bullshit guide for GPU computing on Linux!


 
©2022 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.