Hi,
during compilation of SDK the following error occurred:
nvcc fatal : Unsupported gpu architecture ‘compute_10’
…/…/common/common.mk:445: die Regel für Ziel „obj/x86_64/release/MonteCarlo_SM10.cu_10.o“ scheiterte
make[2]: *** [obj/x86_64/release/MonteCarlo_SM10.cu_10.o] Fehler 1
make[2]: Verzeichnis „/home/marcel/NVIDIA_GPU_Computing_SDK/C/src/MonteCarlo“ wird verlassen
Makefile:43: die Regel für Ziel „src/MonteCarlo/Makefile.ph_build“ scheiterte
make[1]: *** [src/MonteCarlo/Makefile.ph_build] Fehler 2
make[1]: Verzeichnis „/home/marcel/NVIDIA_GPU_Computing_SDK/C“ wird verlassen
Makefile:3: die Regel für Ziel „all“ scheiterte
make: *** [all] Fehler 2
System: Ubuntu 16.04
Nvidia: GeForce GTX 460
Cuda 8.0
Any idea?
The NVIDIA_GPU_Computing_SDK is old, deprecated, no longer shipping, no longer supported. It was shipped up to CUDA 4.2 (about 6 years ago). It has been replaced by the CUDA samples, which get installed with the CUDA 8.0 toolkit.
If you want to compile the old NVIDIA_GPU_Computing_SDK under CUDA 8.0, there will be a lot of work to do to bring it up-to-date. For example you will need to modify compilation targets to remove those that no longer exist, like sm_1x
The following posting will generally cover some of the steps needed if you want to build the (old, deprecated, no longer shipping, no longer supported) NVIDIA_GPU_Computing_SDK using a newer CUDA toolkit:
[url]https://devtalk.nvidia.com/default/topic/827637/cuda-programming-and-performance/mars-mapreduce-framework-amp-cuda-toolkit-7-0/[/url]
The following posting may also be of interest:
[url]Where to find CUDA's cutil_math.h? - Stack Overflow
I would recommend if possible trying to use the newer CUDA samples (like the ones that get installed with CUDA 8) rather than trying to use content from the old SDK.