Welcome to MilkyWay@home

Computation Error on every GPU WU


Advanced search

Message boards : Number crunching : Computation Error on every GPU WU
Message board moderation

To post messages, you must log in.

AuthorMessage
thaumielx72
Avatar

Send message
Joined: 26 Mar 08
Posts: 8
Credit: 729,573
RAC: 0
500 thousand credit badge10 year member badge
Message 46354 - Posted: 25 Feb 2011, 22:53:48 UTC

Hello.

I am running the latest stable version of Ubuntu. (64 bits) Everything is updated and the system is quite stable.

I have recently (within the last few days) loaded the hardware Drivers to reflect a brand new Nvidia GTX 480. (Overclocked, with special state of the art cooling system)

BOINC will load Wu after Wu as Milkyway@Home 0.54 (cuda_opencl) every one of these will error out immediately with Computation Error.

Note: This WAS happening on my Windows partition as well until I updated to the latest Nvidia graphics drivers, now they run smooth as silk.

Windows is OK, but I would love to spend most of my time on Ubuntu Linux.

Has anyone had any luck getting CUDA to run on Ubuntu 64?

Thanks in advance.

*** Just for grins - I went from 150,000 MW credits accumulated over the last few years to 175,000 OVERNIGHT with the Windows CUDA running non-stop (Whew!) ***
ID: 46354 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileKeith Myers
Avatar

Send message
Joined: 24 Jan 11
Posts: 253
Credit: 119,801,841
RAC: 72,776
100 million credit badge8 year member badgeextraordinary contributions badge
Message 46362 - Posted: 26 Feb 2011, 7:21:12 UTC - in response to Message 46354.  
Last modified: 26 Feb 2011, 7:23:00 UTC

I am running the 0.54 Linux OpenCL app on 64 bit Ubuntu 10.10 with the Nvidia 260.19.36 driver with no apparent issues. There were several work units from about a week ago that were formulated wrong and they produced invalid results and computation errors but they should have been flushed from the system by now. Which version of the Nvidia driver are you using? Have you looked in the message log for the errors? Are you running an anonymous app_info?

Cheers, Keith
ID: 46362 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
thaumielx72
Avatar

Send message
Joined: 26 Mar 08
Posts: 8
Credit: 729,573
RAC: 0
500 thousand credit badge10 year member badge
Message 46369 - Posted: 26 Feb 2011, 16:53:33 UTC - in response to Message 46362.  

I am running the 0.54 Linux OpenCL app on 64 bit Ubuntu 10.10 with the Nvidia 260.19.36 driver with no apparent issues. There were several work units from about a week ago that were formulated wrong and they produced invalid results and computation errors but they should have been flushed from the system by now. Which version of the Nvidia driver are you using? Have you looked in the message log for the errors? Are you running an anonymous app_info?

Cheers, Keith


Hmmm. Nvidia X Server says my current driver is 195.36.24. I would have thought that it would load out the most recent driver available.

I used the System > Hardware Drivers menu command. I'm betting your method of loading drivers is more sophisticated.

Well, let me prowl the Ubuntu forums for a bit..

(and thanks!)
ID: 46369 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileKeith Myers
Avatar

Send message
Joined: 24 Jan 11
Posts: 253
Credit: 119,801,841
RAC: 72,776
100 million credit badge8 year member badgeextraordinary contributions badge
Message 46370 - Posted: 26 Feb 2011, 18:55:42 UTC - in response to Message 46369.  

Hi, I would bet that your very old drivers are the reason for the comp errors. I think that the latest MW apps are meant to be used with minimum 258.xxx series drivers. Use the Synaptic Manager or apt-get install to get the latest stable release which is 260.19.36. I wouldn't try the beta 270.xxx series yet through the PPA which I did try and had very poor results and switched back to the stable release. Check the installed and latest columns in Synaptic Manager to see if the 260.19.36 driver appears or if necessary use the Force Update feature of the Manager. You can also directly download the 260.19.36 drivers from Nvidia's website and install through them through a root shell after exiting from the X server.

Keith
ID: 46370 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
thaumielx72
Avatar

Send message
Joined: 26 Mar 08
Posts: 8
Credit: 729,573
RAC: 0
500 thousand credit badge10 year member badge
Message 46371 - Posted: 27 Feb 2011, 0:10:14 UTC - in response to Message 46370.  

You were absolutely correct, sir!

Going to Nvidia's website and running the script by hand was the trick. That, and the use of the Coolbits utility has me humming along in Ubuntu Linux.

Thank you for your help.
ID: 46371 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileKeith Myers
Avatar

Send message
Joined: 24 Jan 11
Posts: 253
Credit: 119,801,841
RAC: 72,776
100 million credit badge8 year member badgeextraordinary contributions badge
Message 46372 - Posted: 27 Feb 2011, 0:48:43 UTC - in response to Message 46371.  

Great! I'm glad that you were able to get the latest Nvidia stable release drivers going. I infer that you are happily crunching away on MW WU's without errors now?

Keith
ID: 46372 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
thaumielx72
Avatar

Send message
Joined: 26 Mar 08
Posts: 8
Credit: 729,573
RAC: 0
500 thousand credit badge10 year member badge
Message 46377 - Posted: 27 Feb 2011, 12:11:53 UTC

Without errors to say the least...

My recent average credit is over 7,000. In less than 2 weeks I may effectively DOUBLE my credits from the last 4 years. To say that I am crunching numbers would be like saying Hawkeye Pierce is telling jokes.

All these things, while fun, are also very good.
ID: 46377 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileKeith Myers
Avatar

Send message
Joined: 24 Jan 11
Posts: 253
Credit: 119,801,841
RAC: 72,776
100 million credit badge8 year member badgeextraordinary contributions badge
Message 46381 - Posted: 28 Feb 2011, 0:31:21 UTC

I know what you mean. I started crunching back in 2001 and had just gone past 200K in December. I finally got around to building myself a new computer and in the last month since beginning GPU computing, I have now gone past 1.2M credits. Incredible to say the least!

Keith
ID: 46381 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Computation Error on every GPU WU

©2019 Astroinformatics Group