nvcc fatal : Unsupported gpu architecture 'compute_20'

[994374821]

If this is a build issue, please fill out the template below.

System information

• Operating system: Ubuntu 16.04
• CMake version: 3.11.0
• Relevant libraries/versions (e.g. CUDA):
• CUDA version : 9.1
• CuDNN version : 7.1.1
• I compile caffe2 with anaconda2

CMake summary output

******** Summary ********
-- General:
--   CMake version         : 3.11.0
--   CMake command         : /home/gaomingda/anaconda2/envs/my_caffe2_env/bin/cmake
--   Git version           : v0.8.1-1499-g409e96d-dirty
--   System                : Linux
--   C++ compiler          : /usr/bin/c++
--   C++ compiler version  : 5.4.0
--   BLAS                  : Eigen
--   CXX flags             : -D__GLIBCXX_USE_CXX11_ABI=0 -Wno-deprecated -fvisibility-inlines-hidden -DONNX_NAMESPACE=onnx_c2 -O2 -fPIC -Wno-narrowing -Wno-invalid-partial-specialization
--   Build type            : Release
--   Compile definitions   : 
-- 
--   BUILD_BINARY          : ON
--   BUILD_CUSTOM_PROTOBUF : OFF
--     Protobuf compiler   : /home/gaomingda/anaconda2/envs/my_caffe2_env/bin/protoc
--     Protobuf includes   : /home/gaomingda/anaconda2/envs/my_caffe2_env/include
--     Protobuf libraries  : /home/gaomingda/anaconda2/envs/my_caffe2_env/lib/libprotobuf.so;-lpthread
--   BUILD_DOCS            : OFF
--   BUILD_PYTHON          : ON
--     Python version      : 3.6.5
--     Python includes     : /home/gaomingda/anaconda2/envs/my_caffe2_env/include/python3.6m
--   BUILD_SHARED_LIBS     : ON
--   BUILD_TEST            : ON
--   USE_ATEN              : OFF
--   USE_ASAN              : OFF
--   USE_CUDA              : ON
--     CUDA version        : 9.1
--     CuDNN version       : 7.1.1
--     CUDA root directory : /usr/local/cuda-9.1
--     CUDA library        : /usr/lib/x86_64-linux-gnu/libcuda.so
--     CUDA NVRTC library  : /usr/local/cuda-9.1/lib64/libnvrtc.so
--     CUDA runtime library: /usr/local/cuda-9.1/lib64/libcudart.so
--     CUDA include path   : /usr/local/cuda-9.1/include
--     NVCC executable     : /usr/local/cuda-9.1/bin/nvcc
--     CUDA host compiler  : /usr/bin/cc
--   USE_EIGEN_FOR_BLAS    : 1
--   USE_FFMPEG            : OFF
--   USE_GFLAGS            : ON
--   USE_GLOG              : ON
--   USE_GLOO              : ON
--   USE_LEVELDB           : OFF
--   USE_LITE_PROTO        : OFF
--   USE_LMDB              : ON
--     LMDB version        : 0.9.21
--   USE_METAL             : OFF
--   USE_MKL               : 
--   USE_MOBILE_OPENGL     : OFF
--   USE_MPI               : OFF
--   USE_NCCL              : ON
--   USE_NERVANA_GPU       : OFF
--   USE_NNPACK            : ON
--   USE_OBSERVERS         : ON
--   USE_OPENCV            : ON
--     OpenCV version      : 3.3.1
--   USE_OPENMP            : OFF
--   USE_PROF              : OFF
--   USE_REDIS             : OFF
--   USE_ROCKSDB           : OFF
--   USE_ZMQ               : OFF

nvcc fatal   : Unsupported gpu architecture 'compute_20'
CMake Error at gloo_cuda_generated_nccl.cu.o.Release.cmake:215 (message):
  Error generating
  /home/gaomingda/caffe2/build/third_party/gloo/gloo/CMakeFiles/gloo_cuda.dir/nccl/./gloo_cuda_generated_nccl.cu.o

third_party/gloo/gloo/CMakeFiles/gloo_cuda.dir/build.make:77: recipe for target 'third_party/gloo/gloo/CMakeFiles/gloo_cuda.dir/nccl/gloo_cuda_generated_nccl.cu.o' failed
make[2]: *** [third_party/gloo/gloo/CMakeFiles/gloo_cuda.dir/nccl/gloo_cuda_generated_nccl.cu.o] Error 1
CMakeFiles/Makefile2:743: recipe for target 'third_party/gloo/gloo/CMakeFiles/gloo_cuda.dir/all' failed

How to solve this problem? Thanks!

[EvrenGursoy]

Is the Cuda and CuDNN version you installed supported by your GPU? If you have an older GPU, you may need to install earlier versions of these.