It is well-known that CUDA toolkit uses LLVM backend, but the used version number is not shown. We can use gdb and LLVM API function to print the version string:
$ gdb /opt/cuda/nvvm/bin/cicc
(gdb) start
(gdb) p 'llvm::cl::PrintVersionMessage'()
Low Level Virtual Machine [http://llvm.org/](http://llvm.org/):
llvm version 3.0
Optimized build.
Built Mar 13 2014 (11:31:40).
Host: i386-pc-linux-gnu
Host CPU: i686
So, CUDA 6.0 uses LLVM 3.0. Current LLVM stable release is 3.4.2.
Dmitry Mikushin
If you need help with Machine Learning, Computer Vision or with GPU computing in general, please reach out to us at Applied Parallel Computing LLC.