Whenever I try to compile with Nsight Eclipse in release mode it takes like 40 minutes. Has anyone else experienced the same problem?
Nsight Eclipse Edition does not introduce any overhead over the command-line nvcc compiler. It simply generates a makefile and then runs make utility that is a part of the OS distribution.
It is likely these long compilation times are caused by the sheer code size and heavy optimizations CUDA compiler does in the release mode. One common reason for the slowdown may be that you have selected all possible CUDA hardware capabilities (SM versions). In this case, compiler will completely recompile your GPU code for every hardware capability level.
To verify the hardware capability levels, right-click your Nsight project, select Properties, select Build/CUDA in the left-hand list. Only check those PTX and GPU code options that you would like to have in your code. Please note that PTX is forward compatible so you may only want to select the earliest version that your application supports. GPU code is not compatible between major releases.
I have never experienced the long compilation time before compiling with Nsight and CUDA 5. In windows with Visual studio there is no problem. I only compile for SM 3.0.
Can you delete the Release folder (the one where compiled binaries are put), rebuild the project and paste here the build log (e.g. the contents of the Console view)?
Are you the same guy who reported the [CBS] bug #192272?
Yes Nvidia managed to reproduce the problem with my code and are now trying to figure out why it is so slow.
Any progress with this bug?