尽管 Cuda 8.0 和带有 GTX 1060 的 nvidia-driver 375 仍无法安装 Torch

Question

我目前正在尝试在我的 Ubuntu 16.04 机器上安装 torch。我已经安装了 Cuda 8.0，nvidia-smi 为我提供了我正在使用的 375.66 版本。

但是，我在尝试运行时收到以下错误消息

./install.sh

从~/torch 文件夹中。

-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/lbfgs.lua
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/checkgrad.lua
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/adamax.lua
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/ConfusionMatrix.lua
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/adadelta.lua
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/rmsprop.lua
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/logger.md
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/intro.md
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image/parameterflattening.svg.png
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image/parameterflattening.svg
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image/parameterflattening.png
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/algos.md
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/logger_plot.png
-- Installing: /home/yenicelik/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/README.md
Updating manifest for /home/yenicelik/torch/install/lib/luarocks/rocks
optim 1.0.5-0 is now built and installed in /home/yenicelik/torch/install/ (license: BSD)

Found CUDA on your machine. Installing CUDA packages
Warning: unmatched variable LUALIB

jopts=$(getconf _NPROCESSORS_CONF)

echo "Building on $jopts cores"
cmake -E make_directory build && cd build && cmake .. -DLUALIB= -DLUA_INCDIR=/home/yenicelik/torch/install/include -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS} -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="/home/yenicelik/torch/install/bin/.." -DCMAKE_INSTALL_PREFIX="/home/yenicelik/torch/install/lib/luarocks/rocks/cutorch/scm-1" && make -j$jopts install

Building on 4 cores
-- Found Torch7 in /home/yenicelik/torch/install
CMake Error at /usr/share/cmake-3.5/Modules/FindPackageHandleStandardArgs.cmake:148 (message):
  Could NOT find CUDA (missing: CUDA_CUDART_LIBRARY) (found suitable version
  "7.5", minimum required is "6.5")
Call Stack (most recent call first):
  /usr/share/cmake-3.5/Modules/FindPackageHandleStandardArgs.cmake:388 (_FPHSA_FAILURE_MESSAGE)
  /home/yenicelik/torch/install/share/cmake/torch/FindCUDA.cmake:1009 (find_package_handle_standard_args)
  CMakeLists.txt:7 (FIND_PACKAGE)


-- Configuring incomplete, errors occurred!
See also "/home/yenicelik/torch/extra/cutorch/build/CMakeFiles/CMakeOutput.log".
See also "/home/yenicelik/torch/extra/cutorch/build/CMakeFiles/CMakeError.log".

我认为这是一个非常简单的错误，但我不确定它为什么会失败。

Answer 1

答案可能对其他人有所帮助。我的神经因此而混乱。 所以，

• 可能在系统中安装了 libcudart7.5。寻找这种参考。
• 如果没有从 /usr/local/cuda-8-0 到 /usr/local/cuda 建立一个链接（sudo ln -s from to)。
• 使用 ./clean.sh 文件，或者更暴力的方式，销毁 torch 目录并克隆一个新目录。出现错误后不会重建引用。
• 安装并享受。

Answer 2

我遇到了同样的错误。只是：

sudo rm /usr/bin/nvcc

并使用您在 /usr/local/cuda-*** 中手动安装的 cuda