Welcome to MilkyWay@home

v0.82 (ati14) computation error

Message boards : Number crunching : v0.82 (ati14) computation error
Message board moderation

To post messages, you must log in.

AuthorMessage
Vampus

Send message
Joined: 6 Mar 08
Posts: 7
Credit: 30,471,731
RAC: 1,333
Message 52002 - Posted: 18 Dec 2011, 5:32:20 UTC

Windows 7 - 64bit
v0.82 (ati14)

For the past couple of weeks my v0.82 (ati14) WUs have been receiving computation errors after 1-2 seconds. I've had a few hundred results like this. I have tried resetting the project, but the results continue.
ID: 52002 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[boinc.at] Nowi

Send message
Joined: 22 Mar 09
Posts: 99
Credit: 503,422,495
RAC: 0
Message 52003 - Posted: 18 Dec 2011, 7:28:56 UTC - in response to Message 52002.  

Can you please tell us something more, like the driver version you use? Have you problems with other BOINC-GPU-Projects? etc.
ID: 52003 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 52010 - Posted: 18 Dec 2011, 21:25:14 UTC - in response to Message 52002.  

There seem to be issues with recent drivers on older cards. Try downgrading to Catalyst 11.8.
ID: 52010 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vampus

Send message
Joined: 6 Mar 08
Posts: 7
Credit: 30,471,731
RAC: 1,333
Message 52057 - Posted: 26 Dec 2011, 7:24:12 UTC - in response to Message 52010.  

Thanks for the replies. I finally find some more info on another thread. My card is a HD4870 with the most current drivers.

Matt: Is this something that will be patched in the next release?
ID: 52057 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Who Know's

Send message
Joined: 10 Nov 09
Posts: 20
Credit: 30,395,419
RAC: 0
Message 52071 - Posted: 26 Dec 2011, 21:48:08 UTC - in response to Message 52057.  

i have a ati 4870 32 bit win7 box and i am running ver 11.5 driver with out any problems
ID: 52071 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,950,829
RAC: 21,429
Message 52085 - Posted: 27 Dec 2011, 16:40:39 UTC - in response to Message 52057.  

Thanks for the replies. I finally find some more info on another thread. My card is a HD4870 with the most current drivers.

Matt: Is this something that will be patched in the next release?


My understanding is that AMD made a major change to its software and the change is not helpful to us crunchers but is to gamers, their biggest market. The bits and pieces I have seen at other projects says that there is a major change coming in the Server side software coming which could talk a long time. Easiest thing is too downgrade to version 11.9 or older and keep on crunching. Here is 'the fix' according to the Admin over at Collatz: "The "fix" would be removing the Brook+ app and installing a much slower OpenCL version which means fewer credits for you and less work done for the project. I am hoping AMD gets their OpenCL performance issues straightened out before they cease all CAL/Brook+ support." This affects the AMD 4XXX series cards the most right now.
ID: 52085 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vampus

Send message
Joined: 6 Mar 08
Posts: 7
Credit: 30,471,731
RAC: 1,333
Message 52102 - Posted: 28 Dec 2011, 6:36:02 UTC - in response to Message 52085.  

My understanding is that AMD made a major change to its software and the change is not helpful to us crunchers but is to gamers, their biggest market.


Unfortunately, I straddle both of those fences :) Guess I might need to rotate out of MW for a while to give both sides a chance to catch up on this.
ID: 52102 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,950,829
RAC: 21,429
Message 52104 - Posted: 28 Dec 2011, 13:07:27 UTC - in response to Message 52102.  

My understanding is that AMD made a major change to its software and the change is not helpful to us crunchers but is to gamers, their biggest market.


Unfortunately, I straddle both of those fences :) Guess I might need to rotate out of MW for a while to give both sides a chance to catch up on this.


AND to give US a chance to catch up to YOU!!!
ID: 52104 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vampus

Send message
Joined: 6 Mar 08
Posts: 7
Credit: 30,471,731
RAC: 1,333
Message 52107 - Posted: 28 Dec 2011, 17:53:48 UTC - in response to Message 52104.  

My understanding is that AMD made a major change to its software and the change is not helpful to us crunchers but is to gamers, their biggest market.


Unfortunately, I straddle both of those fences :) Guess I might need to rotate out of MW for a while to give both sides a chance to catch up on this.


AND to give US a chance to catch up to YOU!!!


Well true, I am awesome. But with a single machine crunching (and having produced nothing but computation errors lately), I'm no threat to the top 1000, let alone top 100 :)
ID: 52107 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 52242 - Posted: 5 Jan 2012, 23:56:03 UTC - in response to Message 52085.  

Here is 'the fix' according to the Admin over at Collatz: "The "fix" would be removing the Brook+ app and installing a much slower OpenCL version which means fewer credits for you and less work done for the project. I am hoping AMD gets their OpenCL performance issues straightened out before they cease all CAL/Brook+ support." This affects the AMD 4XXX series cards the most right now.
The next release now that the server is up will replace the CAL application with OpenCL. However I've done some hackery so that the actual IL kernel that runs is essentially unchanged and is as fast as the old one.
ID: 52242 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Tex1954

Send message
Joined: 22 Apr 11
Posts: 64
Credit: 899,280,380
RAC: 7,829
Message 52246 - Posted: 6 Jan 2012, 1:21:26 UTC
Last modified: 6 Jan 2012, 1:22:30 UTC

Win-7 64b, XFX HD6990, AMD 955BE 3.9GHz

BOINC version 7.0.3, ATI 12.1 Beta Catalyst Drivers

Using this setup, some similar (sometimes) errors at start were corrected.

I discovered 7.0.3 is enforcing the CPU load thing. So, when I run Milkyway, it adds up the 0.80 CPU load and disables one core. A look at Task Manager shows this is not required since it is only 75% loaded.

I have no problems running any other tasks, but I used to be able to run Milkyway 0.82 with all 4 cores crunching other tasks.

Now, when I run Milkyway, I loose 25% of my crunching capability.

Could someone fix the 0.82 (ati) tasks so the CPU's is set to 0.08 instead of 0.80?

Thanks!

8-)

Thanks!
ID: 52246 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 52283 - Posted: 7 Jan 2012, 19:39:51 UTC - in response to Message 52246.  


Could someone fix the 0.82 (ati) tasks so the CPU's is set to 0.08 instead of 0.80?
Fixed
ID: 52283 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Ray_GTI-R
Avatar

Send message
Joined: 5 Nov 10
Posts: 69
Credit: 15,064,831
RAC: 0
Message 52299 - Posted: 7 Jan 2012, 22:07:52 UTC - in response to Message 52002.  

v0.82 (ati14) For the past couple of weeks my v0.82 (ati14) WUs have been receiving computation errors after 1-2 seconds. I've had a few hundred results like this. I have tried resetting the project, but the results continue.

Just started getting tasks failing at 100% complete with the message "Computation error". See http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=61984376 and others for more details.

CCC 11.8 Driver 88.81 (AGP) HD3850 XP Pro SP3
ID: 52299 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 52302 - Posted: 8 Jan 2012, 0:26:18 UTC - in response to Message 52299.  

v0.82 (ati14) For the past couple of weeks my v0.82 (ati14) WUs have been receiving computation errors after 1-2 seconds. I've had a few hundred results like this. I have tried resetting the project, but the results continue.

Just started getting tasks failing at 100% complete with the message "Computation error". See http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=61984376 and others for more details.

CCC 11.8 Driver 88.81 (AGP) HD3850 XP Pro SP3
I would try just rebooting that one to see if it goes away.
ID: 52302 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Ray_GTI-R
Avatar

Send message
Joined: 5 Nov 10
Posts: 69
Credit: 15,064,831
RAC: 0
Message 52304 - Posted: 8 Jan 2012, 1:28:16 UTC - in response to Message 52302.  
Last modified: 8 Jan 2012, 1:31:41 UTC

v0.82 (ati14) For the past couple of weeks my v0.82 (ati14) WUs have been receiving computation errors after 1-2 seconds. I've had a few hundred results like this. I have tried resetting the project, but the results continue.

Just started getting tasks failing at 100% complete with the message "Computation error". See http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=61984376 and others for more details.

CCC 11.8 Driver 88.81 (AGP) HD3850 XP Pro SP3
I would try just rebooting that one to see if it goes away.

Thanks for that.
Yep, the reboot worked. 1 mid-flight task failed as before, then the next few worked OK.

I'd rather not have to do this on a random/regular basis nor waste electricity processing tasks to 100% completion for them to then fail. Can you share what the problem is?
ID: 52304 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Len LE/GE

Send message
Joined: 8 Feb 08
Posts: 261
Credit: 104,050,322
RAC: 0
Message 52305 - Posted: 8 Jan 2012, 4:23:28 UTC

Your repeated error was:

Failed to map resource: Operational error (CAL_RESULT_ERROR)
Failed to release CAL resource


So it looked like your GPU/driver needed a fresh start.
Did you do the reboots when deinstalling 12.1 and installing 11.8?
ID: 52305 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Nolion

Send message
Joined: 6 Nov 11
Posts: 1
Credit: 3,202,614
RAC: 0
Message 52707 - Posted: 28 Jan 2012, 7:25:59 UTC - in response to Message 52242.  

Here is 'the fix' according to the Admin over at Collatz: "The "fix" would be removing the Brook+ app and installing a much slower OpenCL version which means fewer credits for you and less work done for the project. I am hoping AMD gets their OpenCL performance issues straightened out before they cease all CAL/Brook+ support." This affects the AMD 4XXX series cards the most right now.
The next release now that the server is up will replace the CAL application with OpenCL. However I've done some hackery so that the actual IL kernel that runs is essentially unchanged and is as fast as the old one.

To get the POEM@Home OpenCL application running I had to use Catalyst 11.12, and I get the MilkyWay computation errors discussed on this thread. Linux x86_64, BOINC 7.0.2, HD4850. Previously MW ran well on this hardware under Windows with older Catalyst. Looking forward to trying the ATI OpenCL app!
ID: 52707 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
joe

Send message
Joined: 29 Oct 11
Posts: 2
Credit: 9,287,014
RAC: 0
Message 52800 - Posted: 4 Feb 2012, 7:31:22 UTC

Hello,

I have an Ati 4870 and the same problem.

The last driver who work perfectly is this one

11-10_vista64_win7_64_dd_ccc_ocl.exe.

All other after this one failed with v0.82 (ati14).

So when I update catalyst, I do a selective update and uncheck the video driver update.
ID: 52800 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : v0.82 (ati14) computation error

©2024 Astroinformatics Group