Value 'sm_13' is not defined for option 'gpu-architecture'
Issue #68
new
make fails while compiling pyca. Please let me know what am doing wrong :(
Comments (2)
-
-
Hello, I had the same problem yesterday. did you run this code ?
git checkout v0.01
I ran it and got this issue, but i noticed the NOTE given down the instruction:
NOTE: the current version (0.01) is very old, until 0.02 comes out you are probably better off using the development head, just omit the previous steps.
The author tells us to omit the previous steps, and I tried again then I succeed.
May this helps.
- Log in to comment
The default GPU target architecture is set to Compute Capability 1.3 for whatever (old-ish) version of PyCA you're trying to build. Support for compute capabilities < 3.0 have since been removed from the current CUDA toolkit and nvcc compiler by Nvidia.
You can set the target architecture to the version appropriate for your card with the CMake variable
CUDA_ARCH_VERSION
. This variable can be set in either the CMake GUI, or with ccmake on the command line after entering 'advanced mode' by pressingt
. Do this before runningmake
. Do not include the decimal point. Find the compute capability for your GPU here: https://developer.nvidia.com/cuda-gpusFor example, for the GeForce GTX 1080, set the
CUDA_ARCH_VERSION
to 61. Anything greater than or equal to 30 (that is a valid compute capability) should work with CUDA 9.1 (the current CUDA release at time of writing).Alternatively, you could download/install a sufficiently old CUDA relase.