I wanted to update from cuda 3.2 to cuda 4.0, because i needed to retrieve maximum number of resident threads per multiprocessor programmatically.
I found out that the 64 bit 4.0 JCudpp is missing.
As said on the download page there is no Cudpp in Cuda 4.0, but anyway if there is no new dll, then the old one should be in the package at least, because the documentations and jar files are also in there!
http://www.jcuda.de/downloads/JCuda-All-0.4.0-beta1-bin-windows-x86_64.zip
JCudpp-windows-x86_64.dll is missing, so for now I replaced it with the one from 3.2 hope that will work.