CUDA version Problem

thank you very much! i will try to set RTK_CUDA_PROJECTIONS_SLAB_SIZE=32 and rebuild again.

4090

A4000


it seems that there is no change,
here is my cmake setting:

i have tried set RTK_CUDA_PROJECTIONS_SLAB_SIZE=64,but it doesn’t work,the timing information changes just a little bit .

Something is not as expected, you still have 23 calls to the ramp filter and the backprojection. Have you recompiled RTK and your software?

yes,i am sure i have recompiled RTK and software,and replace the new dlls to the software,you mean if change the setting and recompile RTK,ramp filter doesn’t have 23 calls?

image
i have tried to recompile ITK and RTK,in ITK compiling,i notice this setting,it means use NVidia cuda fft with it’s fftw interface for fft computation,but when i choose it,compiling will show error,and there is no ITKFFT-5.4.dll,according to timing information,the problem may lies here.but how can i fix the compiling error?

To use RTK, you might not even need FFTW, much less its CUDA backend. I suggest to keep that option OFF, unless you are trying to improve it.

Each call treats RTK_CUDA_PROJECTIONS_SLAB_SIZE projections at a time so modifying it should modify the number of calls. I don’t know what could be wrong here.

Correct, ITK_USE_CUFFTW should be useless here. This option has been added after I had implemented the Cuda version of the ramp filter.

1 Like

ok.i got it,thank you!