Very good throuhput 98-99% gpu load 31% memory load even with enabled cpu tasks. 135k credits for 11,3 hours on 570 cards. Great units :)

---

DSKAG Austria Research Team: http://www.research.dskag.at

I12R14-GIANNI_VIL1-0-3-RND0846_1
GeForce 660 Ti, Win7 x64
GPU Usage: 70% (!)
GPU Power: 77%
Disabling all CPU tasks raises GPU Usage to 74%, that's still very low.
2.65% completed after 10 minutes
That gives ~6.5hr runtime. Quite much for a short task with such low credit...

Yes, apparently since we used the old simulation files they were configured to produce the old file type. I apologize. Tomorrow morning I will probably cancel the rest and resend them with the new smaller file type.

To cancel units, which we have already started crunching and been crunching for hours, which were otherwise good, is not right and is just a wasting our time. You should have let those units already crunching finish, and correct this problem on all future units.




http://www.gpugrid.net/workunit.php?wuid=4463666

http://www.gpugrid.net/workunit.php?wuid=4463657

http://www.gpugrid.net/workunit.php?wuid=4463672

100% agree..

---

DSKAG Austria Research Team: http://www.research.dskag.at

100% agree too. It was actually unintentional and won't be repeated.

I think you can selectively server abort tasks that have not started.
If a task isn't going to be used for research purposes, there is no point running it. So a server abort of running tasks makes sense, so long as they are marked as valid and credit assigned in full or according to run time.

I12R14-GIANNI_VIL1-0-3-RND0846_1
GeForce 660 Ti, Win7 x64
GPU Usage: 70% (!)
GPU Power: 77%
Disabling all CPU tasks raises GPU Usage to 74%, that's still very low.
2.65% completed after 10 minutes
That gives ~6.5hr runtime. Quite much for a short task with such low credit...

I'm presently running two GIANNI_VIL1 WU's on a 660 and a 660Ti (W7):
With 5 CPU WU's also running my GPU usage is ~84% and 81%
With 2 CPU WU's also running my GPU usage is ~87% and 84%
With no CPU WU's running my GPU usage is ~90% and 87%

Memory Controller loads are 31% and 25% and GDDR memory usage is 160MB and 225MB.

Under my setup the GIANNI_VIL1 WU's are 22% faster on the GTX660Ti than the GTX660 (that was with 5CPU WU's running).

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

100% agree too. It was actually unintentional and won't be repeated.

Very bad form to cancel WUs that have started:

http://www.gpugrid.net/workunit.php?wuid=4453202

http://www.gpugrid.net/workunit.php?wuid=4453531

Over 23 hours wasted :-(

Guys, let me step in. Cancelling WUs was supposed to be conservative (unsent only). If there is a bug - sorry about that, it never surfaced before. I'm having a look.

Yes, apparently since we used the old simulation files they were configured to produce the old file type. I apologize. Tomorrow morning I will probably cancel the rest and resend them with the new smaller file type.

To cancel units, which we have already started crunching and been crunching for hours, which were otherwise good, is not right and is just a wasting our time. You should have let those units already crunching finish, and correct this problem on all future units.




http://www.gpugrid.net/workunit.php?wuid=4463666

http://www.gpugrid.net/workunit.php?wuid=4463657

http://www.gpugrid.net/workunit.php?wuid=4463672

Agreed. One unit finished fine in 8 hours. The other was progressing and was cancelled after 5 hours. Not happy.

http://www.gpugrid.net/workunit.php?wuid=4453678

These things happen in IT as long as lessons are learned it's no biggie - Much better to cancel WUs which can't be used for science than letting hundreds of unsuspecting volunteers globally use thousands of kWh just to allow for credits.

I do however think communication could have been handled better - I would have expected an official thread for this issue with an apology from the senior project manager - In my opinion it has been handled very unprofessionally tarnishing volunteer confidence in the project.

We are here to help the science. BOINC was made to use computing capacity of not used computers for an amount of time. During dinner or something.
Seems to me that we are all "demanding work" from the scientists to be happy crunchers.
If WU's have errors, bugs or no (longer) use for the scientific goal than to me it is absolutely logic to cancel them as soon as possible.

Should be fair for the "credit hunters" to give credits for the run time though.

---

Greetings from TJ

He did say he was going to cancel them in this thread where we were talking about upload size. There wasn't a bug, it was the upload size he was fixing and he unintentionally cancelled sent work units while canceling the wu's in the server. Everybody makes mistakes and he apologized for it.