I am a beginner in both CUDA, English.
I will be able to read English, but to write it is not good.
So, this article has also been automatically translated.
I’ve created a CUDA program.
It seems to work well with GTX-580@CUDA5.0 and -arch sm_13 option.
However, it is not working well with GTX-Titan@CUDA5.5 and -arch sm_35 option.
Calculation accuracy is getting worse about two orders of magnitude.
It is set to double precision mode in nvidia-settings.