nvcc fatal : Value 'sm_50' is not defined for option 'gpu-architecture'


I had GPU rendering with CUDA working on my machine, but after buying a new graphics card, it’s broken. My system summary:

OS: Debian GNUL/Linux testing branch 64-bit
Blender version: 2.70a
nvcc version: release 5.5, V5.5.0
Nvidia driver version: 340.24
Graphics card: GeForce GTX 750 Ti

The error message on the console is:

CUDA version 5.5 detected, build may succeed but only CUDA 5.0 is officially supported.
Compiling CUDA kernel ...
"nvcc" -arch=sm_50 -m64 --cubin "/usr/share/blender/scripts/addons/cycles/kernel/kernel.cu" -o "/home/pelle/.config/blender/2.70/cache/cycles_kernel_sm50_22564524875218C3A66481C514C64EE5.cubin" --ptxas-options="-v"  -I"/usr/share/blender/scripts/addons/cycles/kernel" -DNVCC -D__KERNEL_CUDA_VERSION__=55
nvcc fatal   : Value 'sm_50' is not defined for option 'gpu-architecture'
CUDA kernel compilation failed, see console for details.

Refer to the Cycles GPU rendering documentation for possible solutions:

The error message seems to imply that my new graphics card is not supported by CUDA 5.5. On the other hand, this page seems to imply that it should be supported already in 5.0 (or does “Compute Capability” refer to something different than CUDA version?).

There is also the warning message that only CUDA 5.0 is officially supported, but I’m not very eager to uninstall all Debian supplied Nvidia packages and install the older stuff manually, so I want to hear first if you really think that that could be the problem?

As I read the release notes of Blender 2.71, the answer appeared under my nose:

Added support for Nvidia Maxwell cards (sm_50).

And after upgrading it now works again.