nvcc fatal : Unsupported gpu architecture 'compute_60'

[Jeremy Rutman]

a script that has until now been fine for compiling caffe is now running into the aforementioned error. Anyone have  a clue? There are 2x K80 gpus on the machine. 

# Install Caffe
git clone -b master --depth 1 https://github.com/BVLC/caffe.git /root/caffe
cd /root/caffe
cat python/requirements.txt | xargs -n1 pip install
mkdir build && cd build
cmake -DUSE_CUDNN=1 -DBLAS=Open -DBUILD_python=ON -DBUILD_python_layer=ON ..
....

nvcc fatal   : Unsupported gpu architecture 'compute_60'

details from the cmake:

-- ******************* Caffe Configuration Summary *******************

-- General:

--   Version           :   1.0.0-rc3

--   Git               :   365ac88

--   System            :   Linux

--   C++ compiler      :   /usr/bin/c++

--   Release CXX flags :   -O3 -DNDEBUG -fPIC -Wall -Wno-sign-compare -Wno-uninitialized

--   Debug CXX flags   :   -g -fPIC -Wall -Wno-sign-compare -Wno-uninitialized

--   Build type        :   Release

-- 

--   BUILD_SHARED_LIBS :   ON

--   BUILD_python      :   ON

--   BUILD_matlab      :   OFF

--   BUILD_docs        :   ON

--   CPU_ONLY          :   OFF

--   USE_OPENCV        :   ON

--   USE_LEVELDB       :   ON

--   USE_LMDB          :   ON

--   ALLOW_LMDB_NOLOCK :   OFF

-- 

-- Dependencies:

--   BLAS              :   Yes (Open)

--   Boost             :   Yes (ver. 1.54)

--   glog              :   Yes

--   gflags            :   Yes

--   protobuf          :   Yes (ver. 2.5.0)

--   lmdb              :   Yes (ver. 0.9.10)

--   LevelDB           :   Yes (ver. 1.15)

--   Snappy            :   Yes (ver. 1.1.0)

--   OpenCV            :   Yes (ver. 2.4.8)

--   CUDA              :   Yes (ver. 7.5)

-- 

-- NVIDIA CUDA:

--   Target GPU(s)     :   Auto

--   GPU arch(s)       :   sm_20 sm_21 sm_30 sm_35 sm_50 sm_60 sm_61

--   cuDNN             :   Yes (ver. 5.1.3)

-- 

-- Python:

--   Interpreter       :   /usr/bin/python2.7 (ver. 2.7.6)

--   Libraries         :   /usr/lib/x86_64-linux-gnu/libpython2.7.so (ver 2.7.6)

--   NumPy             :   /usr/lib/python2.7/dist-packages/numpy/core/include (ver 1.8.2)

-- 

-- Documentaion:

--   Doxygen           :   No

--   config_file       :   

-- 

-- Install:

--   Install path      :   /root/caffe/build/install

-- 

-- Configuring done

-- Generating done

-- Build files have been written to: /root/caffe/build

[Przemek D]

It seems that you're trying to use CUDA 7.5 to compile code for devices with compute capability 6.0 and 6.1, which are unsupported.
You have to either switch to 8.0 (perhaps you do have it installed but for some reason cmake chooses to use the old 7.5 one?), OR disable compilation for sm_60 and sm_61. There are two reasons you might want to do both: 1) I see you want to use cuDNN 5.1.3 which AFAIR requires CUDA 8; 2) Tesla K80 has compute capability 3.7, so using architectures above sm_50 is useless (only increases your compilation times).