How to install the supplied caffe with CUDA / cuDNN support?

idata · ‎09-26-2017

Hi,

I want to use the same machine for training, tuning and prototyping. For efficient training caffe needs to use CUDA / cuDNN, but the toolkit installs without it. If I comment out this line in setup.sh

# sed -i 's/CPU_ONLY "Build Caffe without CUDA support" OFF/CPU_ONLY "Build Caffe without CUDA support" ON/' CMakeLists.txt

cmake correctly installs with CUDA / cuDNN but make fails afterwards:

[  0%] Running C++/Python protocol buffer compiler on /opt/movidius/caffe/src/caffe/proto/caffe.proto
Scanning dependencies of target proto
[  0%] Building CXX object src/caffe/CMakeFiles/proto.dir/__/__/include/caffe/proto/caffe.pb.cc.o
[  1%] Linking CXX static library ../../lib/libproto.a
[  1%] Built target proto
[  1%] Building NVCC (Device) object src/caffe/CMakeFiles/cuda_compile.dir/layers/cuda_compile_generated_crop_layer.cu.o
nvcc fatal   : Unsupported gpu architecture 'compute_61'
CMake Error at cuda_compile_generated_crop_layer.cu.o.cmake:207 (message):
  Error generating
  /opt/movidius/caffe/build/src/caffe/CMakeFiles/cuda_compile.dir/layers/./cuda_compile_generated_crop_layer.cu.o


src/caffe/CMakeFiles/caffe.dir/build.make:469: recipe for target 'src/caffe/CMakeFiles/cuda_compile.dir/layers/cuda_compile_generated_crop_layer.cu.o' failed
make[2]: *** [src/caffe/CMakeFiles/cuda_compile.dir/layers/cuda_compile_generated_crop_layer.cu.o] Error 1
CMakeFiles/Makefile2:304: recipe for target 'src/caffe/CMakeFiles/caffe.dir/all' failed
make[1]: *** [src/caffe/CMakeFiles/caffe.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: *** [all] Error 2
Installation failed. Error on line 129

Is it possible to install the supplied caffe with CUDA / cuDNN support or do I need to have an 'external' installation of it?

idata · ‎09-26-2017

Easiest way would be to have caffe-gpu installed in a separate virtualenv. Just remember to modify your PYTHONPATH and exclude movidius out of it when running caffe-gpu.