I've had CUDA Toolkit 9.0 correctly installed on Ubuntu 16.04 with the package manager-based method since both the two commands
cat /proc/driver/nvidia/version
nvcc -V
returne the right infomation. However, it fails to make correctly when I tried to compile the official example NVIDIA_CUDA-9.0_Samples.
io@msi:~/NVIDIA_CUDA-9.0_Samples$ make
The main output is
/usr/bin/ld: can't find -lcudart
collect2: error: ld returned 1 exit status
Makefile:334: recipe for target 'simpleMPI' failed
make[1]: *** [simpleMPI] Error 1
make[1]: Leaving directory '/home/io/NVIDIA_CUDA-9.0_Samples/0_Simple/simpleMPI'
Makefile:52: recipe for target '0_Simple/simpleMPI/Makefile.ph_build' failed
make: *** [0_Simple/simpleMPI/Makefile.ph_build] Error 2
can anyone give me a hand? Thank you.
I think I've solved this problem. As I mentioned in the question, I've had CUDA Toolkit correctly installed in my computer, but it fails when execute
io@msi:~/NVIDIA_CUDA-9.0_Samples$ make
However, the following two commands
$ ~/NVIDIA_CUDA-9.0_Samples/1_Utilities/deviceQuery/deviceQuery
$ ~/NVIDIA_CUDA-9.0_Samples/1_Utilities/bandwidthTest/bandwidthTest
gives correct response "Result = pass". It's ok to ignore the "make error" listed above- I think.