Error: Unknown CUDA Architecture Name 2.1(2.0) in CUDA_SELECT_NVCC_ARCH_FLAGS #665
Description
I tried to build pytorch from source code on Ubuntu 14.04 using the cuda release 8.0. But it failed to build.
Error:
-- Autodetected CUDA architecture(s): 2.1(2.0)
CMake Error at /home/sudar/pytorch/cmake/FindCUDA/FindCUDA/select_compute_arch.cmake:142 (message):
Unknown CUDA Architecture Name 2.1(2.0) in CUDA_SELECT_NVCC_ARCH_FLAGS
Call Stack (most recent call first):
CMakeLists.txt:115 (CUDA_SELECT_NVCC_ARCH_FLAGS)
CMake Error at /home/sudar/pytorch/cmake/FindCUDA/FindCUDA/select_compute_arch.cmake:146 (message):
arch_bin wasn't set for some reason
Call Stack (most recent call first):
CMakeLists.txt:115 (CUDA_SELECT_NVCC_ARCH_FLAGS)
-- got cuda version 8.0
-- Found CUDA with FP16 support, compiling with torch.CudaHalfTensor
-- CUDA_NVCC_FLAGS: -DTH_INDEX_BASE=0 -I/home/sudar/pytorch/torch/lib/tmp_install/include -I/home/sudar/pytorch/torch/lib/tmp_install/include/TH -I/home/sudar/pytorch/torch/lib/tmp_install/include/THC;-DCUDA_HAS_FP16=1
-- THC_SO_VERSION: 1
-- Configuring incomplete, errors occurred!
See also "/home/sudar/pytorch/torch/lib/build/THC/CMakeFiles/CMakeOutput.log".