After attaching to Seti@home using Boinc Manager version 7.0.64 and Windows Ultimate 7, 32 bits, I had 18 CUDA tasks completed which resulted in an error before I noticed that I once again had forgot to set the CUDA_GRID_SIZE_COMPAT environment variable and therefore had to restart Boinc Manager once again.

This environment variable is still needed for my GTX 680 4 GB cards, I assume.

The nVidia graphics driver version is now 314.22, by the way.

If I am not wrong, forgetting to set this environment variable leads to error 12.

Why did many of these tasks result in error 1 instead? Is the card (or really graphics driver) working and functional when getting this error, or is this perhaps a computational error which I may receive at other times when the environment variable has already been set?

Thanks for explanation!

...I noticed that I once again had forgot to set the CUDA_GRID_SIZE_COMPAT environment variable and therefore had to restart Boinc Manager once again.

This environment variable is still needed for my GTX 680 4 GB cards, I assume.

The nVidia graphics driver version is now 314.22, by the way.

If I am not wrong, forgetting to set this environment variable leads to error 12.
...

No, error -12 is a data-dependant error indication meaning "Unsupported function". When NVIDIA ported the code to CUDA, it was judged that data leading to two triplets in one array or more than 10 bins above threshold in a triplet array would be so rare that allocating additional Video RAM to allow for those cases wasn't sensible.

Yes, the GTX 680 is a Kepler card and needs the environment variable to successfully run the NVIDIA ported CUDA applications.

                                                                  Joe

Thanks Joe.

Have a look at http://setiathome.berkeley.edu/workunit.php?wuid=1243943382 for example.

This task is having a little longer run time length than some others when it comes to my error task (Task 2994419633).

Error 1 is apparently reported there, but possibly the driver was already functional (perhaps even before the environment variable had been set).

In fact I think I should be able to tell when the driver is not working correctly.

Or am I perhaps still misreading it once again?

*spock* [For several reasons]

Looks like slight differences in startup time to me - CPU time is mostly higher as well.
A Kepler/stock setup will only work with the variable set.
Tasks that error out straight away almost always point to some problem with the setup (e.g. wrong app, bad driver, not enough memory...)

---

A person who won't read has no advantage over one who can't read. (Mark Twain)

Thanks Joe.

Have a look at http://setiathome.berkeley.edu/workunit.php?wuid=1243943382 for example.

This task is having a little longer run time length than some others when it comes to my error task (Task 2994419633).

Error 1 is apparently reported there, but possibly the driver was already functional (perhaps even before the environment variable had been set).

In fact I think I should be able to tell when the driver is not working correctly.

Or am I perhaps still misreading it once again?

The error shown in the Stderr output for that workunit is, "UFFT error in file 'd:/Projects/SETI/seti_boinc/client/cuda/cudaAcc_fft.cu' in line 62.", which shows that the environment variable had not been set.

Did you restart the PC after setting the environment variable?

Cheers.

Back again and making a closer check, definitely error 1 is the CUDA_GRID_SIZE_COMPAT environment variable not having been set error message while error 12 is the "cudaAcc_find_triplets erroneously found a triplet twice in find_triplets_kernel" error message.

Therefore this problem has now been solved and interpreted by me.

The only thing which actually confused me here was the indication message in the result lines for both types of errors stating that both GTX 680 cards were working (for both error 1 and error 12).

But apparently not as they really should because of the environment variable not having been set where error 1 is present.