Computation Errors


log in

Advanced search

Questions and Answers : Windows : Computation Errors

Author Message
Silvas
Send message
Joined: 23 Jan 01
Posts: 11
Credit: 2,685,897
RAC: 1,878
United States
Message 1352563 - Posted: 31 Mar 2013, 20:50:40 UTC
Last modified: 31 Mar 2013, 21:00:03 UTC

My old machine expired and I have replaced with an I7-3750k and it uses an Nvidia GTX680 grafix card. System is Windows 7 and whatever the current version of Boinc for the past month being available.

After setting this up to run Seti@home and Einstein@home which I ran on the old machine, everything appeared to be working fine. When I looked in Tasks I saw that every seti file had a computation error rather than completed. All the Einstein files run through with nary a problem.

The above paragraphs tell you everything I know about this operation. I am not a techie and it's probably a miracle than I've managed to accomplish anything with Seti in the last dozen years. Currently I have suspended Seti (I think)so that I won't be submitting corrupted files ..however I haven't a clue as how to fix this and I don't find anything in any of the other messages posted applying ..at least that I can understand. Help would be appreciated but think of giving instruction at a Dick, Jane, and Spot level.

The version of boinc is 7.0.28 (x64) that I have operating.
____________

Profile Ageless
Avatar
Send message
Joined: 9 Jun 99
Posts: 12341
Credit: 2,639,943
RAC: 1,145
Netherlands
Message 1352599 - Posted: 31 Mar 2013, 22:14:57 UTC - in response to Message 1352563.

See point 2 in this sticky thread in the GPU forum.

2) Kepler cards unsupported

Drivers affected: All drivers from 304.48 (BETA) onwards.
Hardware affected: nVidia 'Kepler' - GT 6xx and GTX 6xx cards.

Symptom: All tasks end in errors when using the stock v6.10 'cuda_fermi' application.

Solution/workround:
a) Use an optimised Cuda application, where available.
b) Downgrade to the 301.42 driver - only possible on GTX 670/680/690 cards, not on newer releases like the 650/660 or their Ti variants.
c) Set an environment variable, as shown below.

The environment variable to use is

CUDA_GRID_SIZE_COMPAT

and it needs to have the value 1 (one)

You need to be running in Administrator mode, and you need to be running at least driver 306.02 (BETA) or 306.23 (WHQL)

____________
Jord

Fighting for the correct use of the apostrophe, together with Weird Al Yankovic

Silvas
Send message
Joined: 23 Jan 01
Posts: 11
Credit: 2,685,897
RAC: 1,878
United States
Message 1353732 - Posted: 5 Apr 2013, 16:27:39 UTC - in response to Message 1352599.

Firstly, I have no idea what an "optimized cuda application" is nor do I understand why this should be necessary. Downgrade my driver(s)? I don't think so.. they're working beautifully with the rest of my system and I was always told to not "fix" something that isn't broken. Obviously it is the Boinc client that newer technology has broken, not my driver. And I have not learned how to "set an environment variable". Apparently gone are the good ol' days when all one had to do was run the seti client app and set a few limitations in the options. If there are no plans for Boinc to be updated to make it work with all our video cards then I suppose it will be time for me to cut this machine loose from Seti. I have an older machine that is running and working well with Seti (the one my wife uses) and will have to be satisfied with allowing it to run until such time as it is updated or becomes toast.

Sorry, I'm not going to jump through hoops to harness my system to Seti's specifications so if Seti/Boinc doesn't like what I have to offer (freely given and willingly) then I suppose it will have to do without. And dear reader, there are a lot of us who are not computer gurus that are willing to contribute to the effort but simply wish to do so hassle free and without problems from an unruly and/or underdeveloped client app.

____________

Profile Ageless
Avatar
Send message
Joined: 9 Jun 99
Posts: 12341
Credit: 2,639,943
RAC: 1,145
Netherlands
Message 1353843 - Posted: 5 Apr 2013, 20:03:11 UTC - in response to Message 1353732.
Last modified: 5 Apr 2013, 20:10:17 UTC

And I have not learned how to "set an environment variable".

Then go to the FAQ that I pointed out (the blue link in my earlier post), which will explain how you do that. It's really easy, no need to know the internal workings of BOINC or Windows. They even use pictures. Edit: ah heck, here it is:


For your information, it isn't BOINC that is broken, but the Seti application. BOINC does not do any of the calculations, that's done by the project's science applications. The application that you're getting sent by the Seti servers is updated for Fermi style cards, not for the Kepler style cards, such as you have. Thus the application will use the hardware 'wrong', which results in computation errors.

The environment variable is an easy workaround against the problem.
____________
Jord

Fighting for the correct use of the apostrophe, together with Weird Al Yankovic

Silvas
Send message
Joined: 23 Jan 01
Posts: 11
Credit: 2,685,897
RAC: 1,878
United States
Message 1354263 - Posted: 7 Apr 2013, 6:11:39 UTC - in response to Message 1353843.

I am grateful for your time and assistance, thank you. And the information you have passed along is noted ...I did follow the instructions to set an environmental variable. I actually fault no one but myself for not taking the time to become more computer literate, however time has been an ungrateful b*ch and seems to come only in small slices. I hope someone in Seti finds the wherewithal to pay the application programmer a little more so that all of us can just run it and go on about our business without these hairpulling sessions.

Again, let me thank you for your time and effort.

____________

Silvas
Send message
Joined: 23 Jan 01
Posts: 11
Credit: 2,685,897
RAC: 1,878
United States
Message 1354423 - Posted: 7 Apr 2013, 16:14:24 UTC - in response to Message 1353843.

Btw... merely setting the variable was not a "fix" for the problem.. so I suppose all three (a,b and c) in that instructional commentary must be done to achieve the workaround.

Thanks again.. enjoyed while it lasted. On to Einstein@home.

____________

Profile Ageless
Avatar
Send message
Joined: 9 Jun 99
Posts: 12341
Credit: 2,639,943
RAC: 1,145
Netherlands
Message 1354424 - Posted: 7 Apr 2013, 16:26:56 UTC - in response to Message 1354423.
Last modified: 7 Apr 2013, 16:27:06 UTC

No, it's or, or, or, not and, and, and. Can't even be at this time, as the optimized application has compiler copyright issues, so it's not available at this time.

But you'd be better off posting about it in the Number Crunching forum, in a new thread. Do point out this thread, so people won't ask you to jump through the same hoops twice. The optimized app people mainly post over there.
____________
Jord

Fighting for the correct use of the apostrophe, together with Weird Al Yankovic

Questions and Answers : Windows : Computation Errors

Copyright © 2014 University of California