Advanced search

Message boards : News : No longer cuda3.1 apps in acemdlong queue.

Author Message
Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1957
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 27916 - Posted: 2 Jan 2013 | 13:27:24 UTC

As announced we have deprecated the cuda3.1 application for the acemdlong queue.
The acemd2 queue stays with both applications.

From now on you should not be receiving acemdlong cuda3.1 apps.

gdf

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27920 - Posted: 2 Jan 2013 | 21:10:34 UTC - in response to Message 27916.
Last modified: 2 Jan 2013 | 21:13:13 UTC

From now on you should not be receiving acemdlong cuda3.1 apps.

Houston, we have a problem.

potocnyj
Send message
Joined: 5 Aug 12
Posts: 1
Credit: 66,839,315
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwat
Message 27921 - Posted: 3 Jan 2013 | 6:06:51 UTC - in response to Message 27920.

Your task says it was created on Dec 19th, so it should just be a leftover. Or were all existing 3.1 tasks deleted from the queue? If they were not deleted, it may be wise to remove them so that we do not end up crunching tasks that will not be used for research purposes.

Bedrich Hajek
Send message
Joined: 28 Mar 09
Posts: 467
Credit: 8,189,021,966
RAC: 10,556,536
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27922 - Posted: 3 Jan 2013 | 8:47:04 UTC

Here is another problem:


http://www.gpugrid.net/workunit.php?wuid=4014336

http://www.gpugrid.net/result.php?resultid=6236277



Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27925 - Posted: 3 Jan 2013 | 10:42:52 UTC - in response to Message 27916.
Last modified: 3 Jan 2013 | 10:54:54 UTC

Perhaps a project reset is required to push deprecation?
The tasks can run on either the 3.1 or 4.2app, so perhaps Boinc is controlling which app the tasks runs on locally, meaning we would need to delete the 3.1 app locally, with a project reset or an app update?
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 27927 - Posted: 3 Jan 2013 | 11:01:53 UTC - in response to Message 27925.

BTW even if crunched with previous cuda, results are still useful for research.

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1957
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 27929 - Posted: 3 Jan 2013 | 11:10:34 UTC - in response to Message 27927.

I did a project restart.
Let me know.

gdf

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27932 - Posted: 3 Jan 2013 | 14:57:31 UTC - in response to Message 27929.
Last modified: 4 Jan 2013 | 21:53:38 UTC

I had a look in the Boinc Projects folder and the 3.1 dll files were there.
After a project update these 3.1 files remained, but after Resetting the project they are no longer there, and hopefully wont be downloaded again.
AFAIK the tasks are sent in 3.1 or 4.2 format, so Resetting the project is not necessary, and certainly not recommended when a task is running.
The newly named app will automatically download, because the server has assigned the tasks to run on that app. It's shouldn't therefore run on the 3.1app, even if you have it.
Anyone wanting to run tasks on the 3.1app should only crunch tasks from the normal length queue, and not the Long task queue.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Profile dskagcommunity
Avatar
Send message
Joined: 28 Apr 11
Posts: 456
Credit: 817,865,789
RAC: 0
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27933 - Posted: 3 Jan 2013 | 16:03:23 UTC

So it could be a problem when a cuda42 capable machine is in both queues? O.o
____________
DSKAG Austria Research Team: http://www.research.dskag.at



Profile Stoneageman
Avatar
Send message
Joined: 25 May 09
Posts: 224
Credit: 34,057,224,498
RAC: 231
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27934 - Posted: 3 Jan 2013 | 18:09:26 UTC

Just wasting the electricity equivalent of the cost of a premium beer today crunching three 3.1 tasks which slipped through. No doubt the boinc schedulers personality is a killjoy.

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1957
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 27937 - Posted: 4 Jan 2013 | 11:17:46 UTC - in response to Message 27934.

I have now created app 6.17 for cuda4.2.
It's the same application of 6.16, just a way to convince the server that cuda31 is deprecated.

gdf

Profile Chilean
Avatar
Send message
Joined: 8 Oct 12
Posts: 98
Credit: 385,652,461
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27939 - Posted: 4 Jan 2013 | 12:37:18 UTC - in response to Message 27937.

...just a way to convince the server...


This made me chuckle.

Jorge Alberto Ramos Olive...
Send message
Joined: 13 Aug 09
Posts: 24
Credit: 156,684,745
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwat
Message 27983 - Posted: 9 Jan 2013 | 13:26:49 UTC

Anyone care to explain what 3.1 or 4.2 is? Is it the CUDA driver version?

and now that I'm into it: what about compute capability? Most recent cards (6xx) are 3.0, how important is this version to GPUGrid?

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27994 - Posted: 9 Jan 2013 | 17:32:40 UTC - in response to Message 27983.

Your questions have been covered in other threads, but I'll answer them anyway. 3.1 and 4.2 are applications that use CUDA 3.1 and 4.2 respectively. So I often refer to them as the 3.1app and 4.2app. It's actually down to the .dll files more than the CUDA version, but lets not split hairs.
The 301.42 driver is sufficient to support the CUDA 4.2 application, however some newer GeForce 600 GPU's require a more recent driver (or at least say they do).

Compute capability (CC) is just a description of each GPU's mathematical genre. These include CC1.0 (doesn't work here), CC1.1 (probably wont work), CC1.2 (might still work badly), CC1.3 (no longer efficient), CC2.0 (Fermi), CC2.1 (different generation of Fermi on which only 2/3rds of the shaders are usable for here), CC3.0 (Kepler and again only 2/3rds of the shaders are usable for here).

The CC describes the GPU's abilities, and the researchers use this to determine the GPU type. The top CC2.0 GPU's (Geforce GTX 590, 580, 570, 480, 560 Ti (448), 470, 465) are less power efficient than the top CC3.0 GPU's (GTX690, GTX680, GTX670, GTX660Ti, GTX660, GTX650Ti, GTX650) making the CC3.0 cards preferable. The GTX570 and GTX580 are roughly as powerful as the GTX660Ti, GTX670 and GTX680, but being less power efficient and an older generation they are slightly down the recommended list.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Nikke
Send message
Joined: 25 Aug 10
Posts: 9
Credit: 2,143,920,937
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28029 - Posted: 14 Jan 2013 | 9:32:03 UTC - in response to Message 27994.

Dear all,

could anyone explain me why I'm getting only 3.1 normal length jobs?
Some info:

CentOS Linux 5.5
BOINC 6.10.58
CUDA Runtime 4.2.9
Driver 310.19 (just upgraded from 295.41 to test if it was the problem)

Already tried projects reset and update.....

Some hints, things to check?

Thanks in advance

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28032 - Posted: 14 Jan 2013 | 12:00:46 UTC - in response to Message 28029.

Probably because all your long 4.2 tasks failed,
http://www.gpugrid.net/results.php?hostid=78927&offset=0&show_names=0&state=0&appid=23
This might have been related to your driver. I would suggest doing a project restart and restarting your system. Then select the long queue (or both) and see how you get on.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Nikke
Send message
Joined: 25 Aug 10
Posts: 9
Credit: 2,143,920,937
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28036 - Posted: 14 Jan 2013 | 14:55:02 UTC - in response to Message 28032.

Probably because all your long 4.2 tasks failed,
http://www.gpugrid.net/results.php?hostid=78927&offset=0&show_names=0&state=0&appid=23
This might have been related to your driver. I would suggest doing a project restart and restarting your system. Then select the long queue (or both) and see how you get on.

Unfortunately, even after a reboot and a project reset and changing prefs to get only long jobs I'm currently not able to fetch any job.

The client reports something like "No tasks are available for the applications you have selected"

Could it be related to the quite old version of BOINC I'm running? Or to the OS?
Sorry, I'm not so good in debugging BOINC tasks...

Thanks,

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28039 - Posted: 14 Jan 2013 | 15:38:23 UTC - in response to Message 28036.
Last modified: 14 Jan 2013 | 16:40:57 UTC

So you have set your Project preferences to choose Long tasks,
http://www.gpugrid.net/prefs.php?subset=project
and presumably made sure the profile was correct for your machine.

It can take a while to start getting tasks again, but I see you have more 3.1 tasks from the 'normal' length queue.

Perhaps you need additional library files?
sudo apt-get install ia32-libs libstdc++6 libstdc++5 freeglut3

Might also be a good idea to update Boinc, especially after a driver update,
apt-get install boinc-client boinc-manager

--
I see you are now getting this error message with the Long tasks,
../../projects/www.gpugrid.net/acemd.2562.x64.cuda42: /lib64/libc.so.6: version `GLIBC_2.7' not found (required by ./libcudart.so.4)

You have a CentOS library problem.

Came across this possible solution,
Run find / -name /lib/libc.so*

Should output something like /lib/libc.so.4 or /lib/libc.so.5 or /lib/libc.so.6

Make a soft link, ln -s /lib/libc.so.4 /lib/libc.so.6.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1957
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 28047 - Posted: 14 Jan 2013 | 19:54:30 UTC - in response to Message 28039.

This would not probably work.

You need to install the cuda4.2 toolkit for redhat5 and set in .bashrc LD_LIBRARY_PATH to point to /usr/lib64/cuda

We distribute redhat6 libraries with the app, so you need to override our libs.

gdf

Nikke
Send message
Joined: 25 Aug 10
Posts: 9
Credit: 2,143,920,937
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28052 - Posted: 15 Jan 2013 | 7:54:06 UTC - in response to Message 28047.

I perfectly agree, it is a problem of libraries.
Now I'm trying to solve it or I'm going to build the BOINC client from scratch

Reinstalling the server is not feasible right now

Thanks

chuckbens
Send message
Joined: 15 Oct 12
Posts: 18
Credit: 27,916,718
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 28189 - Posted: 24 Jan 2013 | 14:28:05 UTC

I am fairly new to BOINC and even newer to GPU's so please bear with me...

A team member donated his old GTX275 to me a few months ago and I have been using it here ever since. I have 910K credits and was looking for my first Million. Now, I understand there have been some changes and I am no longer getting and WU's.

I do not know what CUDA 3.1 or CUDA 4.2 is. I have figured out it has nothing to do with my nVidia driver version though! I am running 306.23 and was unable to upgrade to 310.90...

Is the GTX275 compatible with CUDA 4.2? If so, how do I upgrade CUDA on my machine?

Thanks for any and all help!

Chuck

Profile dskagcommunity
Avatar
Send message
Joined: 28 Apr 11
Posts: 456
Credit: 817,865,789
RAC: 0
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28190 - Posted: 24 Jan 2013 | 14:43:00 UTC

I tried a 285gtx with cuda 4.2 but the performance was very poor. Nearly 50% longer then cuda31. So i would recommend to stay on short tasks.
____________
DSKAG Austria Research Team: http://www.research.dskag.at



chuckbens
Send message
Joined: 15 Oct 12
Posts: 18
Credit: 27,916,718
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 28191 - Posted: 24 Jan 2013 | 14:53:52 UTC

OK, this is weird. I disabled long runs in my GPUGrid preferences and then reset the project in BAM. Then I received a CUDA42 long run WU! I was hoping to still get the short CUDA31 tasks...

http://www.gpugrid.net/workunit.php?wuid=4066922


Chuck

chuckbens
Send message
Joined: 15 Oct 12
Posts: 18
Credit: 27,916,718
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 28192 - Posted: 24 Jan 2013 | 14:55:03 UTC - in response to Message 28190.

Thanks! I didn't realize the short tasks were still using the old CUDA31.

Chuck

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28198 - Posted: 24 Jan 2013 | 19:11:51 UTC - in response to Message 28192.
Last modified: 24 Jan 2013 | 19:14:58 UTC

To prevent receipt of tasks from either queue, when editing your Project Preferences unselect the following,

    If no work for selected applications is available, accept work from other applications?


This is usually not required as the queues are normally loaded with plenty of tasks. However application testing is presently under way.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

chuckbens
Send message
Joined: 15 Oct 12
Posts: 18
Credit: 27,916,718
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 28199 - Posted: 24 Jan 2013 | 19:35:34 UTC - in response to Message 28198.

I forgot to do that. However, I am curious why I am getting a CUDA42 task AND it appears to be running OK...

Profile dskagcommunity
Avatar
Send message
Joined: 28 Apr 11
Posts: 456
Credit: 817,865,789
RAC: 0
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28200 - Posted: 24 Jan 2013 | 20:51:37 UTC

On short you get cuda42 And 31. Only with driver under 301 you get cuda31 only.
____________
DSKAG Austria Research Team: http://www.research.dskag.at



chuckbens
Send message
Joined: 15 Oct 12
Posts: 18
Credit: 27,916,718
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 28207 - Posted: 25 Jan 2013 | 14:48:51 UTC - in response to Message 28200.

OK, if I understand you correctly...

With nVidia driver 306.23 I do have CUDA42. However, from your experience, it will run very slow on my GTX275 and I should stick with CUDA31.

I was not aware that driver 306.23 had CUDA42. So, I need to find a stable driver under 301.xx to run only CUDA31 short runs.

Is that correct? If so, what driver would you recommend?

Thanks and sorry for the Noob questions.

Chuck

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 28208 - Posted: 25 Jan 2013 | 15:45:09 UTC - in response to Message 28207.

OK, if I understand you correctly...

With nVidia driver 306.23 I do have CUDA42. However, from your experience, it will run very slow on my GTX275 and I should stick with CUDA31.

I was not aware that driver 306.23 had CUDA42. So, I need to find a stable driver under 301.xx to run only CUDA31 short runs.

Is that correct?

Yes, this is correct.

If so, what driver would you recommend?

I would recommend the driver 285.58.
There are two more recent drivers (296.10 and 295.73) before 301.42 which are also suitable for you, but they have the monitor sleep bug (e.g. you have to set Windows to never turn off your monitor, or else your tasks will fail when it happens)

Post to thread

Message boards : News : No longer cuda3.1 apps in acemdlong queue.

//