Video driver keeps failing on certain GPU WUs using AMD3870 APU.

Message boards : Number crunching : Video driver keeps failing on certain GPU WUs using AMD3870 APU.
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile AllenIN
Volunteer tester
Avatar

Send message
Joined: 5 Dec 00
Posts: 292
Credit: 58,297,005
RAC: 311
United States
Message 1751961 - Posted: 28 Dec 2015, 2:32:19 UTC
Last modified: 28 Dec 2015, 2:44:07 UTC

I can't seem to figure out what is causing the problem. I have tried a half dozen things and nothing seems to work. The work unit will sometimes finish and sometimes fail. Mostly, anything over 2 hours will fail for sure.
I have all the normal settings with no OC. I have installed the latest Boinc version and the latest optimize version as well. Even when running without the optimization, it still fails in the same way.

It says that the video has failed but has recovered and the WU continues to run with no progress. As much as I hate it, I have to abort it in order to begin another.

I have another system that is running the same processor with no problems at all. I have even installed a new, more powerful power supply in the troublesome machine and it hasn't change a thing.

I'm certainly all ears if anyone has any ideas. BTW, this machine ran everything just fine at one time, but that seems like years ago now.

Thanks, Allen
ID: 1751961 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1751966 - Posted: 28 Dec 2015, 3:44:51 UTC - in response to Message 1751961.  

The problem appears to be with the low end GPU, not the processor. The low end ATI cards have problems with the low angle range tasks. You might be able to help it by increasing the -period_iterations_num setting in the mb_cmdline...txt file located in the setiathome.berkeley.edu folder. As per the readme;
-period_iterations_num N: Splits single PulseFind kernel call to N calls for longest PulseFind calls. Can be used to reduce GUI lags or to prevent driver restarts. Can affect performance. Experimentation required. Default value for v6/v7 task is N=20. N should be positive integer.

Try opening your mb_cmdline_win...txt file with notepad and pasting in the setting;
-period_iterations_num 64
If that doesn't help try upping it to 100;
-period_iterations_num 100
ID: 1751966 · Report as offensive
Profile AllenIN
Volunteer tester
Avatar

Send message
Joined: 5 Dec 00
Posts: 292
Credit: 58,297,005
RAC: 311
United States
Message 1751976 - Posted: 28 Dec 2015, 5:46:19 UTC - in response to Message 1751966.  

The problem appears to be with the low end GPU, not the processor. The low end ATI cards have problems with the low angle range tasks. You might be able to help it by increasing the -period_iterations_num setting in the mb_cmdline...txt file located in the setiathome.berkeley.edu folder. As per the readme;
-period_iterations_num N: Splits single PulseFind kernel call to N calls for longest PulseFind calls. Can be used to reduce GUI lags or to prevent driver restarts. Can affect performance. Experimentation required. Default value for v6/v7 task is N=20. N should be positive integer.

Try opening your mb_cmdline_win...txt file with notepad and pasting in the setting;
-period_iterations_num 64
If that doesn't help try upping it to 100;
-period_iterations_num 100


TBar,

You really know your stuff. Thank you so much!!
I tried the 64 and it failed again, so I tried the 100 and it failed again...but it acted like it was thinking about it so I tried one more time by setting it to 180 and it paused when it got to the regular place that it would fail and then popped back and forth and finally got past it for good and went all the way to the end. I must be honest though, I have no idea how this works, but I'm very glad that it did.

You mentioned that the cpu that I am using is on the low end. It's the equivalent of a Radeon 6550 and has 400 streaming processors, which while not being a STRONG GPU, it shouldn't have that much trouble with the number crunching should it?

I just bought another APU, but this one is a R7 level GPU. I hope it will do better. BTW, should I assume that the larger a work unit is the higher I might have to move that number?

Thanks again, Allen
ID: 1751976 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1752076 - Posted: 28 Dec 2015, 16:32:57 UTC - in response to Message 1751976.  

Glad to see it finishes a task now. Strange it has the problem where the other one doesn't. I was confused by the SuperSumo name, somehow I assumed it was a discrete card. It appears the SuperSumo is really a BeaverCreek? I can't even find a SuperSumo on the AMD pages. Weird. About as weird as SETI telling me my Mac has a Opteron with 3DNOW. Oh well.

The HD6xxx cards have problems with the Seti@home Multibeam App even my 6850s & 6870 have bad screen lag with the Low Angle range tasks. I can imagine how it is with the lower end 6xxx cards. Not much you can do about it except run AstroPulses when possible. The one GDDR5 HD7750 I have seems to have much less screen lag than the 6 series cards so hopefully the R7 based GPU will do better. Usually you don't see much difference with settings above 100, but I guess your case proves different.
ID: 1752076 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1752077 - Posted: 28 Dec 2015, 16:36:36 UTC
Last modified: 28 Dec 2015, 16:40:19 UTC

I have optimized a Laptop last year with SuperSumo chip.
Yes it needs rather high period_iterations_num values.
But for that price its O.K.
The Lappy cost only 250 bucks.
Kalindi is much worse.


With each crime and every kindness we birth our future.
ID: 1752077 · Report as offensive
Profile AllenIN
Volunteer tester
Avatar

Send message
Joined: 5 Dec 00
Posts: 292
Credit: 58,297,005
RAC: 311
United States
Message 1752091 - Posted: 28 Dec 2015, 17:34:51 UTC - in response to Message 1752076.  
Last modified: 28 Dec 2015, 17:35:33 UTC

Thanks for all the good input. I'm just sorry that I didn't speak up sooner. I was considering that perhaps I didn't have the proper bios settings, but I had been over them dozens of times so I felt as though it should run!

It's great to have knowledgeable people here to rely on from time to time.

Thanks!

BTW, you are correct, Beavercreek.
ID: 1752091 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1752136 - Posted: 28 Dec 2015, 22:12:22 UTC

Along with period iterations one can consider to increase memory buffer for PulseFind via -sbs 256 setting. Default is 64 to fit into lowest memory requirements.
ID: 1752136 · Report as offensive
Profile AllenIN
Volunteer tester
Avatar

Send message
Joined: 5 Dec 00
Posts: 292
Credit: 58,297,005
RAC: 311
United States
Message 1752694 - Posted: 31 Dec 2015, 16:49:18 UTC - in response to Message 1752136.  
Last modified: 31 Dec 2015, 16:54:01 UTC

Along with period iterations one can consider to increase memory buffer for PulseFind via -sbs 256 setting. Default is 64 to fit into lowest memory requirements.


Interesting.
I finally had on WU fail. I reset the -period_iterations_num to 310 and was able to complete it. I should mention that I also OCed the cpu speed from 3.0 to 3.2 and increased the voltage to both the memory and the cpu slightly.

Looks like it's working fine now, but if I should have any more problems, I think I'll try your suggestion above. I also see that there is something called -sbs 256 -cpu_lock. Don't know anything about that, but that's the way I found the file and there is nothing else in the file. Is that the same as what you suggested?

Thanks again for the help.
ID: 1752694 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1752704 - Posted: 31 Dec 2015, 17:29:47 UTC - in response to Message 1752694.  

Along with period iterations one can consider to increase memory buffer for PulseFind via -sbs 256 setting. Default is 64 to fit into lowest memory requirements.


Interesting.
I finally had on WU fail. I reset the -period_iterations_num to 310 and was able to complete it. I should mention that I also OCed the cpu speed from 3.0 to 3.2 and increased the voltage to both the memory and the cpu slightly.

Looks like it's working fine now, but if I should have any more problems, I think I'll try your suggestion above. I also see that there is something called -sbs 256 -cpu_lock. Don't know anything about that, but that's the way I found the file and there is nothing else in the file. Is that the same as what you suggested?

Thanks again for the help.


Yes, it is the same.

I noticed you have set it at 256 already.


With each crime and every kindness we birth our future.
ID: 1752704 · Report as offensive
Profile AllenIN
Volunteer tester
Avatar

Send message
Joined: 5 Dec 00
Posts: 292
Credit: 58,297,005
RAC: 311
United States
Message 1753078 - Posted: 2 Jan 2016, 5:20:38 UTC - in response to Message 1752704.  
Last modified: 2 Jan 2016, 5:25:42 UTC

Along with period iterations one can consider to increase memory buffer for PulseFind via -sbs 256 setting. Default is 64 to fit into lowest memory requirements.


Interesting.
I finally had on WU fail. I reset the -period_iterations_num to 310 and was able to complete it. I should mention that I also OCed the cpu speed from 3.0 to 3.2 and increased the voltage to both the memory and the cpu slightly.

Looks like it's working fine now, but if I should have any more problems, I think I'll try your suggestion above. I also see that there is something called -sbs 256 -cpu_lock. Don't know anything about that, but that's the way I found the file and there is nothing else in the file. Is that the same as what you suggested?

Thanks again for the help.


Yes, it is the same.

I noticed you have set it at 256 already.


Actually it was already set that way when I went to the file. Can the 256 be increased if necessary?
ID: 1753078 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1753141 - Posted: 2 Jan 2016, 9:10:03 UTC - in response to Message 1753078.  

Along with period iterations one can consider to increase memory buffer for PulseFind via -sbs 256 setting. Default is 64 to fit into lowest memory requirements.


Interesting.
I finally had on WU fail. I reset the -period_iterations_num to 310 and was able to complete it. I should mention that I also OCed the cpu speed from 3.0 to 3.2 and increased the voltage to both the memory and the cpu slightly.

Looks like it's working fine now, but if I should have any more problems, I think I'll try your suggestion above. I also see that there is something called -sbs 256 -cpu_lock. Don't know anything about that, but that's the way I found the file and there is nothing else in the file. Is that the same as what you suggested?

Thanks again for the help.


Yes, it is the same.

I noticed you have set it at 256 already.


Actually it was already set that way when I went to the file. Can the 256 be increased if necessary?


Yes, but keep in mind it will increase memory consumption and your GPU only has 512 MB.


With each crime and every kindness we birth our future.
ID: 1753141 · Report as offensive

Message boards : Number crunching : Video driver keeps failing on certain GPU WUs using AMD3870 APU.


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.