Advanced search

Message boards : Graphics cards (GPUs) : Why does CUDA always occupy a CPU's core ?

Author Message
refla
Send message
Joined: 12 Feb 09
Posts: 9
Credit: 385,357
RAC: 0
Level

Scientific publications
watwatwatwat
Message 6627 - Posted: 14 Feb 2009 | 11:51:51 UTC

I find cuda program will occupy a CPU's core when it is running. But Task Manager shows the core occupied by cuda has done nothing until the cuda program terminates.

Why does not BOINC assign other WU, likes LCH, SETI, and so on, to the idle core? Or it's a bug in gpugrid.net ?

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 6628 - Posted: 14 Feb 2009 | 12:22:32 UTC - in response to Message 6627.

Or it's a bug in gpugrid.net ?


No, it's a bug in BOINC 6.4.5. Use 6.5.0 and you should be fine.

MrS
____________
Scanning for our furry friends since Jan 2002

Daniel Neely
Send message
Joined: 21 Feb 09
Posts: 5
Credit: 34,605,213
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 6845 - Posted: 21 Feb 2009 | 22:39:47 UTC - in response to Message 6628.

Since anything later than 6.4.5 is a dev version, how can I know which ones are stable enough to use?

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 6846 - Posted: 21 Feb 2009 | 22:42:40 UTC - in response to Message 6845.

As far as I understood 6.4.5 is also a development version. If that one does not behave well try 6.5.0.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Bender10
Avatar
Send message
Joined: 3 Dec 07
Posts: 167
Credit: 8,368,897
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwatwat
Message 6847 - Posted: 21 Feb 2009 | 22:49:14 UTC - in response to Message 6846.
Last modified: 21 Feb 2009 | 23:00:01 UTC

Is that a platform specific bug..? I don't have that on 6.4.5 and XP64...(on an X2 core).

If you use 6.4.5 you have to make a cc_config file with some goobledy-gook in it to make it see the CPUs and GPUs as needed.

My 6.4.5 must be twisted...It's stock, and runs fine...
____________


Consciousness: That annoying time between naps......

Experience is a wonderful thing: it enables you to recognize a mistake every time you repeat it.

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 6848 - Posted: 21 Feb 2009 | 22:50:27 UTC - in response to Message 6846.

As far as I understood 6.4.5 is also a development version. If that one does not behave well try 6.5.0.


You have it backwards ... 6.4.5 is "recommended" by UCB ... 6.5.0 is development, but is better and more stable ... which is why we recommend it ... in this case "better" because it runs GPU tasks without needed a configuration file like 6.4.5 and earlier do ...

As to Daniel's question ... you ask people you trust ... :)

If you use 6.4.5 you have to make a cc_config file with some goobledy-gook in it to make it see the CPUs and GPUs as needed. 6.5.0 sees the GPUs correctly (Nvidia only) ...

Of course, the next question, do you trust me ... :)


ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 6849 - Posted: 22 Feb 2009 | 1:01:21 UTC - in response to Message 6848.

You have it backwards ... 6.4.5 is "recommended" by UCB


Recommended by UCB? Oh, I missed that one. The 6.4.x seemed such a step backwards compared to 6.3.21, so the idea that these might be official never crossed my mind. And I wondered why GPU-Grid recommended 6.4.5 .. :D

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 6856 - Posted: 22 Feb 2009 | 4:31:51 UTC - in response to Message 6849.

You have it backwards ... 6.4.5 is "recommended" by UCB


Recommended by UCB? Oh, I missed that one. The 6.4.x seemed such a step backwards compared to 6.3.21, so the idea that these might be official never crossed my mind. And I wondered why GPU-Grid recommended 6.4.5 .. :D


Yeah ...

The problem is that the developers, as best as I can tell are not, ahem, real users of BOINC. By that I mean that they don't seem to have that much interest in the operational use of BOINC to do projects. My opinion is derived from there longstanding opposition to many features that would make BOINC easier to use when running BOINC. I mean, multiple select rows did not get into the thing until 5 years after launch. Even now, how many people know about that feature?

For those that use BOINC as start and forget, and there are a lot of people out there that are like that, BOINC has always been adequate, even bad versions like 6.4.5 which does work if you just leave it alone. *BUT*, if you are passionate about your projects, and involved with them, BOINC is very inadequate. Saddest of all BOINC does not meet several of its design goals nor its original vision.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 6866 - Posted: 22 Feb 2009 | 11:45:05 UTC - in response to Message 6856.

I think I know what you mean. Especially the first BOINC versions, when basically everything was still alpha / beta, it could *almost* drive one mad not to have manual overrides. So hard to tell BOINC what it should do in cases where it didn't automatically do it right. However, since a couple of years it basically ran flawlessly for me.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 6895 - Posted: 22 Feb 2009 | 20:05:16 UTC - in response to Message 6866.

I think I know what you mean. Especially the first BOINC versions, when basically everything was still alpha / beta, it could *almost* drive one mad not to have manual overrides. So hard to tell BOINC what it should do in cases where it didn't automatically do it right. However, since a couple of years it basically ran flawlessly for me.


Overall I also have little to complain about with regard to operation, the only problem is that if you do want to customize the operation some, it is very difficult to do so. If you just want to load and forget it, it will run reasonably well with no intervention.

My problem with it is that with the way I run projects and by careful (obsessive?) observation I can see that it is not really running tasks the way that the developers say it should be running tasks ... more importantly, it does not run them the way I would like them to be run.

That said, mostly I do just let it run ...

The problem is that there are 10 kinds of BOINC users ... :)

The kind that knows binary and those that don't ... some of us want a little more direct control over the flow of work through our systems ... yet the developers insist that if they give us those controls the people that will NOT intervene in their systems will use the controls to mess things up ... as if they are not sometimes messed up already ...

But, we are way off topic here ....

Post to thread

Message boards : Graphics cards (GPUs) : Why does CUDA always occupy a CPU's core ?

//