Just to clarify, I absolutely WANT to use official repository.
This give me the possibility to be automatically updated, for example.
Even if I'm 62, I'm not retired; I'm still working and I want to use my precious to make things in order to earn money. I can't follow strange update every time someone make a change somewhere in the linux word.
Evening, saturdays and sundays are for wife, sons, grandchildren and cats. Not necessarily in the same order, on the same day :-) but sometime I spend time for seti, if you like.

So the only thing I need is something working, even if other are faster.

If something is wrong in the official repository, I hope that sooner or later someone will update the repository. Best if sooner. And sooner will be if I don't follow shortcuts but I show the road found to working solutions.


Just as an experiment (that could be interesting for me and for someone else) I should like to try if multiple tasks in sigle GPU can do the difference.


So I repeat ... If some one can refresh my memory abot how to have multiple task on a single GPU, you are welcome.

Told this, of course I thank you, TBar, for all the given and precious help.

---

Greetings Francesco

You have to adjust the following line in your APP_INFO.XML

<count>1.0</count> to <count>0.5</count>

Regards
Tazzduke

Thanks!

---

You are using a third party SETI App, that will never be automatically updated. Since the 750Ti is almost 2 years old, you will probably never need to change the driver in the foreseeable future, except if for some reason it stops working. Those two items will Not stop everything else from being automatically updated.

BTW, you still don't have a working OpenCL. That means the Official OpenCL App, if there ever is one, will Not work. It also means the AstroPulse App will Not work. It's easy to see if you have OpenCL, just look at the top of your event log. Or, just look at your Computer description;
NVIDIA GeForce GTX 750 Ti (2047MB)
It should say; NVIDIA GeForce GTX 750 Ti (2047MB) OpenCL: 1.2

Whatever, you can't say didn't try.

For the rest of you, installing a working Linux is Very simple. It takes about 45 minutes.
All that is required is to Install the OS, Install the nVidia driver, then Install the Berkeley BOINC.

Tbar wrote:

I have wasted Hours trying to get the Repository 352 driver to see OpenCL after Finally getting it to see CUDA. I don't have that problem with the recent drivers downloaded from nVidia.

Did you try running any OpenCL programs other than BOINC client to see if it's the client or the driver package?

TBar wrote:

My question is if there is Ever going to be an Updated Berkeley copy of BOINC. I've tried 7.4.25 and it doesn't work correctly. The Manager doesn't update without running the pointer across the screen. The other versions I've seen on the Web appear to have been compiled to be installed by the Repository verses placed in the Home folder. I suppose I could try compiling my own version, but it would be nice to just download a recent version from Berkeley.

I believe it's a matter of no money, no time. Compiling it yourself isn't hard. Only catch is if you want Manager then you need wxWidgets 3. The one in Ubuntu 14.04 was compiled without wxWebview and isn't compatible with BOINC. 14.10 and later should have compatible wx and LocutusOfBorg's ppa has compatible wx packages for 14.04.

Francesco Forti wrote:

Iztok wrote:

in /etc/rc.d/rc.local:

nvidia-smi
sleep 1
echo "boinc/dobo >/dev/null 2>&1 &" |su - boinc

and seti works even if X11 is not started on this PC.
nvidia-smi loads proper modules.

I don't have any rc.d in /etc
I have a list of rc0.d to rc6.d + rcS.d but no rc.local file

It's not that easy, Iztok is running Slackware, you have Ubuntu. Way too incompatible.

Francesco Forti wrote:

Just to clarify, I absolutely WANT to use official repository.
This give me the possibility to be automatically updated, for example.

Gianfranco Costamagna aka LocutusOfBorg packages BOINC for Ubuntu. If his package archive isn't official enough then you need to run the following command from terminal to get OpenCL detection working:

cd /usr/lib/x86_64-linux-gnu/ && sudo ln -s libOpenCL.so.1 libOpenCL.so

Your understanding of how Ubuntu (and Mint) work may be incorrect. Once the distribution is released package versions are mostly set in stone. Packages are mainly updated only to fix security issue or other serious bugs. There may be some exceptions, Firefox for example is updated fairly often.

Your understanding of how Ubuntu (and Mint) work may be incorrect. Once the distribution is released package versions are mostly set in stone. Packages are mainly updated only to fix security issue or other serious bugs. There may be some exceptions, Firefox for example is updated fairly often.

I get a lof of updates weekly. Two or three each day. Bugs, security but also
improvements.

---

I'm going to play around with Mint a little later and see how the Repository 352 does. I'd expect it's the same as it was last year though. Since I began with Ubuntu I haven't seen any changes to the Older Repositories. Major Bugs usually have a Patch and then they say it will be fixed In The Next Release. I've only seen One driver change and that was with the ATI driver a while back. As far as I know, the drivers in the Older Releases are Never "fixed". Other Bugs aren't fixed either. This one particular BUG kept me away from 14.04 for over a year,
xorg.conf overwritten by booting system. I still have to comment out lines in gpu-manager every time I install a new 14.04 and use ATI cards.

While looking for the two old threads about the problem with driver 352 I ran across this old link, Boinc + Ubuntu + Nvidia = No usable GPUs found Posted 04/22/2014.
There are older links around, but I'm trying to stay semi-current. I found this thread interesting;

I would humbly suggest you un-install those other things, especially that boinc-nvidia-cuda. I don't have it and I'm using cuda just fine.

The next thing you hear;

OK. The Projects now see the GPUs...

I don't install any of those "extras" and My systems work just fine. Currently, All you need is that Official driver from nVidia, the people that make the thing, and BOINC also from the people that make the thing.

Here's the older threads;
No Usable GPU in Linux
Updating GPU drivers in Linux
Note who was telling people to make links in those threads.

Well, I tried the OC setting and that works in Mint 17.3. Adding Option "Coolbits" "8" works as it did in Ubuntu. You would think that GTX 560 could be downclocked.
I then played around with Blender to see if it would output a status file about having CUDA or whatever, no luck. So, if someone knows of a way to detect CUDA, other than with BOINC, let me know.

Then I removed nVidia driver 361.42 and installed the Repository version of 352-updates.
Here it is;

25-Apr-2016 00:31:35 [---] Starting BOINC client version 7.2.33 for x86_64-pc-linux-gnu
25-Apr-2016 00:31:35 [---] log flags: file_xfer, sched_ops, task, sched_op_debug
25-Apr-2016 00:31:35 [---] Libraries: libcurl/7.35.0 OpenSSL/1.0.1f zlib/1.2.8 libidn/1.28 librtmp/2.3
25-Apr-2016 00:31:35 [---] Data directory: /home/tbar/BOINC
25-Apr-2016 00:31:35 [---] No usable GPUs found
25-Apr-2016 00:31:35 [SETI@home] Found app_info.xml; using anonymous platform
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type NVIDIA
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type CUDA
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type CUDA
25-Apr-2016 00:31:35 [SETI@home Beta Test] Found app_info.xml; using anonymous platform
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type CUDA
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type NVIDIA
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type NVIDIA
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type NVIDIA
25-Apr-2016 00:31:35 [---] app version refers to missing GPU type NVIDIA
25-Apr-2016 00:31:35 [SETI@home] Missing coprocessor for task 06ap10ab.7062.24202.15.42.242_1
25-Apr-2016 00:31:35 [SETI@home] Missing coprocessor for task 07my10af.16016.261821.6.33.94_1
blah blah blah....

The Good news is I don't think it trashed any tasks.

So, I went from a perfectly working BOINC to a Dud by just moving to the Mint Repository driver. There is a reason I tell people to use the driver from nVidia.

Anyone have anything they would like me to try before I move back to that Official, Working Driver from nVidia?

So, if someone knows of a way to detect CUDA, other than with BOINC, let me know.

Just run one of the Cuda apps standalone ( e.g. under bench) It will enumerate the detected devices. I imagine OpenCL devices would do similar.

---

"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.

Evening Jason,

Here it is;

01:27:14 (2370): Can't open init data file - running in standalone mode
Cuda error 'Couldn't get cuda device count
' in file 'cuda/cudaAcceleration.cu' in line 138 : unknown error.
setiathome_CUDA: cudaGetDeviceCount() call failed.
setiathome_CUDA: No CUDA devices found
setiathome_CUDA: Found 0 CUDA device(s):
setiathome_CUDA: No SETI@home capabale CUDA GPU found...
Cuda device initialisation retry 1 of 6, waiting 5 secs...

&

01:28:27 (2390): Can't open init data file - running in standalone mode
Not using mb_cmdline.txt-file, using commandline options.
01:28:27 (2390): Can't open init data file - running in standalone mode
WARNING: init_data.xml missing
Error: Getting Platforms. (clGetPlatformsIDs)
BOINC assigns slot on device #0.
WARNING: BOINC failed to provide OpenCL device, using own enumeration abilities
ERROR: OpenCL kernel/call 'clGetDeviceIDs (second call)' call failed (-32) in file ../../src/GPU_lock.cpp near line 1128.
Waiting 30 sec before restart...

I've already tried making links to the CUDA & OpenCL files in /usr/lib/i386-linux-gnu and placing them in /usr/lib, apparently that doesn't work. I'll see if I can find the libraries somewhere else.
The libraries are also in /usr/lib/x86_64-linux-gnu. Making Links, moving the links to /usr/lib, and naming them libcuda.so & libOpenCL.so doesn't work either. I'm getting this deja vu feeling.
Making more links to the files in /usr/lib/x86_64-linux-gnu, moving them to /usr/lib, and naming them libcuda.so.1 & libOpenCL.so.1 Doesn't Work either. I've definitely seen this before...

Yeah no Cuda or OpenCL present there :D

My familiarity with the Linux fine details is on the light side ( I just install nv drivers and it works for me), but looks like whatever packages are involved there, fail to make one or more standard symlinks. Could be accidental/bug, or some deliberate slimming operation similar to some Windows drivers sent out missing bits.

[Note:] Nothing majorly special about the device enumeration in either of those cases ---> If it isn't there it isn't there :)

---

"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.

Incidentally, I've grown quite attached to the Cuda app's tendency to rub your nose in things repeatedly, will probably keep that, LoL.

---

"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.

Doesn't Work either. I've definitely seen this before...

Seems like maybe the nouveau drivers weren't properly blacklisted, or something about the kernel modules failed, such as missing that kernel module management thing (forget the name) so scripts fail.

---

"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.

I think I did get cuda working before by placing the link somewhere else, possibly /lib. But I can't remember exactly. At this point I'm ready to just remove the links I made in usr/lib, reinstall 361.42 and call it. Oh, I can't find nvidia-modprobe in the Package Manager either. I did see 346.xx there though. Perhaps I'll try installing that...

I think I did get cuda working before by placing the link somewhere else, possibly /lib. But I can't remember exactly. At this point I'm ready to just remove the links I made in usr/lib, reinstall 361.42 and call it. Oh, I can't find nvidia-modprobe in the Package Manager either. I did see 346.xx there though. Perhaps I'll try installing that...

Lol, well probably probing all the wrong ways and smashing your head against it until it works will help. It's how I usually do stuff.

---

"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.

Back in action. Well that was fun....Not!
I did get smart though, made a copy of the BOINC folder and made sure networking was Off every time I tested a new setting with the old folder. Then just replaced the folder with the backup, removed the Repository driver, installed 361.42 and;

Mon 25 Apr 2016 04:03:58 AM CDT | | Starting BOINC client version 7.2.33 for x86_64-pc-linux-gnu
Mon 25 Apr 2016 04:03:58 AM CDT | | Data directory: /home/tbar/BOINC
Mon 25 Apr 2016 04:03:58 AM CDT | | CUDA: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version unknown, CUDA version 8.0, compute capability 5.0, 2043MB, 1887MB available, 2082 GFLOPS peak)
Mon 25 Apr 2016 04:03:58 AM CDT | | OpenCL: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 361.42, device version OpenCL 1.2 CUDA, 2043MB, 1887MB available, 2082 GFLOPS peak)
Mon 25 Apr 2016 04:03:58 AM CDT | | OS: Linux: 3.19.0-32-generic

Yes!

So Repo maintainers are twiddling something ?

---

"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.

From what I've seen it's basically the way Juha described it. Once the system has been released the basic Repo doesn't change. If there is an update/BugFix needed they just release a patch and tell you the Next Release will have the Fix incorporated. It keeps you moving forward I suppose. Those drivers in the repo will Not be changed, you'll just have to use the Next release or use a work around. That's the way it seems to me anyway, the only updates you get are security updates and app fixes.

From the links I've seen the Repo driver problems started around 334 and continued through 352. I haven't seen any Repo drivers above 352. I'll have to install Ubuntu 16.04 and see what it has, but, I'm outta partitions. To install 16.04 something will have to go, and I think I'm going to need a New Berkeley BOINC as well.
Oh well, I think it's way past bedtime here.

Makes sense in a concrete 'touchy feely Linuxey' kindof way I suppose. The Problem is now Valve corporation, Vulkan/Khtonos, Google/Android, and gamers, would like the technology to compete on the VR front, and if it can;t do on basic compute then IMO it's likely another gameover in m$'s favour, Needs more awesomeness.

---

"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.

I repeat my solution:

1) boinc from repository (7.2.42 x64)
2) driver from repository (352.63)
3) nVidia Toolkit 7.5 from here: https://developer.nvidia.com/cuda-toolkit

GPU job is non fast but I think that part of my problems are connected with the GPU internal in my Core i7 6700. Maybe some BIOS setting can do something.

---