I just started to do CUDA since I got my computer fixed and I have a CUDA card and now my WU's has all CUDA and no SETI? What gives?

It saw you had a CUDA card so it sent you CUDA WUs by default. You should also get some astropulse WUs.

---

PROUD MEMBER OF Team Starfire World BOINC

I just started to do CUDA since I got my computer fixed and I have a CUDA card and now my WU's has all CUDA and no SETI? What gives?

You could shoulda had no Cuda as you shouldna.......but you Cuda had a cuda so ya just been Cuda'd..........


Enjoy....

---

"Time is simply the mechanism that keeps everything from happening all at once."

I just started to do CUDA since I got my computer fixed and I have a CUDA card and now my WU's has all CUDA and no SETI? What gives?

You could shoulda had no Cuda as you shouldna.......but you Cuda had a cuda so ya just been Cuda'd..........


Enjoy....

That's good!

My CUDA WU's run faster than my SETI WU's for the same credit in other words more work done with CUDA equals more credit with CUDA. A 14 credit WU with CUDA equals about 50 seconds CPU time. I think if I understand my results page.

My computer number is 37936
See http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=1444420
And http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=1602276

My CUDA WU's run faster than my SETI WU's for the same credit in other words more work done with CUDA equals more credit with CUDA. A 14 credit WU with CUDA equals about 50 seconds CPU time. I think if I understand my results page.
...

You need to look at elapsed time, the 50 seconds is the small amount of CPU time needed to feed the GPU. There are plans for BOINC to gather and report actual time used IIRC. That's not yet available, so the only way to tell is to look in the logged or BOINC Manager messages for the start time and finish time of a task.

                                                              Joe

My CUDA WU's run faster than my SETI WU's for the same credit in other words more work done with CUDA equals more credit with CUDA. A 14 credit WU with CUDA equals about 50 seconds CPU time. I think if I understand my results page.
...

You need to look at elapsed time, the 50 seconds is the small amount of CPU time needed to feed the GPU. There are plans for BOINC to gather and report actual time used IIRC. That's not yet available, so the only way to tell is to look in the logged or BOINC Manager messages for the start time and finish time of a task.

                                                              Joe

OK thanks that makes more sense now.

Yeah, well here's the deal. Other threads got it right. Cuda isn't ready for prime time. Every one of my tasks errored out with not enough memory, and that = no credi-to. Worse, instead of the elegent sharing that has allowed me to continue whatever work I have with S@H I now have sparklies after a CUDA computational error. I opted out and will stay out until one of you brave souls logs about 50 hours with no problems. If there's change review comittee in charge they need to reconsider their choices.

---

Yeah, well here's the deal. Other threads got it right. Cuda isn't ready for prime time. Every one of my tasks errored out with not enough memory, and that = no credi-to. Worse, instead of the elegent sharing that has allowed me to continue whatever work I have with S@H I now have sparklies after a CUDA computational error. I opted out and will stay out until one of you brave souls logs about 50 hours with no problems. If there's change review comittee in charge they need to reconsider their choices.

I see many -9 overflows but my guess is they were caused by This task. This is a Very Low Angle Range task, or VLAR (0.1 angle range and below tasks) that if you didn't reboot after this task, can have the effect of causing what the other tasks that followed had, "SETI@Home Informational message -9 result_overflow
NOTE: The number of results detected exceeds the storage space allocated." which is a false result in most cases under these circumstances.

It's been suggested that you use the batch file descrebed in several of the threads here like this one in an attempt to spot these VLAR tasks before they run so they can be aborted.

---

Got same problem in here. My screen gets black from time to time (couple of times a day atm). I've got this information in the right corner "display driver nvlddmkm stopped response, but it works again now". In many cases i have to reeboot my machine. Its because of thaousands of small pixels moving on my screen. This kind of WUs need less than 30 sec of time to be finished (all with errors, all with Spike count: 30, all with same message 'SETI@Home Informational message -9 result_overflow') Any clue how to fix it ???

---

Got same problem in here. My screen gets black from time to time (couple of times a day atm). I've got this information in the right corner "display driver nvlddmkm stopped response, but it works again now". In many cases i have to reeboot my machine. Its because of thaousands of small pixels moving on my screen. This kind of WUs need less than 30 sec of time to be finished (all with errors, all with Spike count: 30, all with same message 'SETI@Home Informational message -9 result_overflow') Any clue how to fix it ???

If you have troubles with CUDA, it's worth to read other threads where all currently know bugs described and solutions are proposed.
Black screen, driver restart, BSoD, overflows - all these symptoms belongs to VLAR/VHAR processing bugs in current CUDA MB version. The solution is simple you just need to run script sorting tasks by AR, abort all tasks with AR known in "bug area" and do happy CUDA crunching for few days. Then refill cache and repeat procedure.
That's all. And it takes much less time to complete than to write endless moanings here on forums.
Look here too:
http://setiathome.berkeley.edu/forum_thread.php?id=51175

The solution is simple you just need to run script sorting tasks by AR, abort all tasks with AR known in "bug area" and do happy CUDA crunching for few days. Then refill cache and repeat procedure.

Where can I find such a script? I have grown tired of aborting frozen tasks, coping with unnecessary overflows, rebooting, etc. I had a look around and found references to that script but not the script itself.

Thanks in advance.

Roberto who got a 9800 GT just to speed up crunching

---

here: http://setiathome.berkeley.edu/beta/forum_thread.php?id=1443&nowrap=true#36281

here: http://setiathome.berkeley.edu/beta/forum_thread.php?id=1443&nowrap=true#36281

Thank you so much. I've worked out my own script, but without that reference it would have been impossible.
Now let's see if my computer can run unattended for more than a few hours.

---

here: http://setiathome.berkeley.edu/beta/forum_thread.php?id=1443&nowrap=true#36281

Thank you so much. I've worked out my own script, but without that reference it would have been impossible.
Now let's see if my computer can run unattended for more than a few hours.

If you made some useful improvements it's worth to publish your own script too - maybe it will help someone... ;)

I have been reading about the CUDA and how fast it is or isn't.
I am undecided but here is a jpg of my task folder

http://seti.pdharris.com/task/task_2009.12.01.jpg

Got same problem in here. My screen gets black from time to time (couple of times a day atm). I've got this information in the right corner "display driver nvlddmkm stopped response, but it works again now". In many cases i have to reeboot my machine. Its because of thaousands of small pixels moving on my screen. This kind of WUs need less than 30 sec of time to be finished (all with errors, all with Spike count: 30, all with same message 'SETI@Home Informational message -9 result_overflow') Any clue how to fix it ???

If you have troubles with CUDA, it's worth to read other threads where all currently know bugs described and solutions are proposed.
Black screen, driver restart, BSoD, overflows - all these symptoms belongs to VLAR/VHAR processing bugs in current CUDA MB version. The solution is simple you just need to run script sorting tasks by AR, abort all tasks with AR known in "bug area" and do happy CUDA crunching for few days. Then refill cache and repeat procedure.
That's all. And it takes much less time to complete than to write endless moanings here on forums.
Look here too:
http://setiathome.berkeley.edu/forum_thread.php?id=51175

THX for Your tips Raistmer. I'm gonna try to sort this out tomorrow. I'll check forum twice before my next question.

---

I have been reading about the CUDA and how fast it is or isn't.
I am undecided ...

There are two kind of WU's sent out.
1st one needs around 9min
2nd needs around 27min

reference: the cpu on my host needs at least twice

---

If you made some useful improvements it's worth to publish your own script too - maybe it will help someone... ;)

Oh, it just sorts the WUs by true_angle and prints out the results. The only difference is that it is written using Unix shell commands (Windows + Cygwin) because I like bash better than DOS.

Now I was wondering whether I can delete the WU files my CUDA devices are not able to process instead of aborting them manually. I don't know how BoincManager behaves in case of missing files.

By the way, I needed to abort all task with angle < 0.012, because CUDA got struck on one unit with angle = 0.012. That made some 30% of all WUs in the last batch I downloaded. I hope I will not be marked as a "bad" user of sort in S&H databases. My computer has been running fine for 18 hours now, thought.

Thanks again for your help!

Roberto

---

Now I was wondering whether I can delete the WU files my CUDA devices are not able to process instead of aborting them manually. I don't know how BoincManager behaves in case of missing files.

It's better that you abort tasks instead of deleting them unless you following the instructions in this post.

By the way, I needed to abort all task with angle < 0.012, because CUDA got struck on one unit with angle = 0.012. That made some 30% of all WUs in the last batch I downloaded. I hope I will not be marked as a "bad" user of sort in S&H databases. My computer has been running fine for 18 hours now, thought.

Those are Very Low Angle Range or VLAR tasks (0.1 angle range and below) and it's understood those of us using CUDA will be aborting them for now. No one will hold it against you.

---