Failed to compile XGBoost-R on Windows 10 with GPU support

#1

Hi all,

I’ve been trying to compile XGBoost-R on Windows 10 with GPU support:

mkdir build
cd build
cmake .. -G"Visual Studio 15 2017 Win64" -DUSE_CUDA=ON -DR_LIB=ON
cmake --build . --target install --config Release

And I hit the following error

Project "C:\Users\Administrator\Desktop\xgboost\build\xgboost.vcxproj" (1) is building "C:\Users\Administrator\Desktop\
xgboost\build\gpuxgboost.vcxproj" (4) on node 1 (default targets).
PrepareForBuild:
  Creating directory "gpuxgboost.dir\Release\".
  Creating directory "C:\Users\Administrator\Desktop\xgboost\build\Release\".
  Creating directory "gpuxgboost.dir\Release\gpuxgboost.tlog\".
InitializeBuildStatus:
  Creating "gpuxgboost.dir\Release\gpuxgboost.tlog\unsuccessfulbuild" because "AlwaysCreate" was specified.
ComputeCustomBuildOutput:
  Creating directory "C:\Users\Administrator\Desktop\xgboost\build\CMakeFiles\gpuxgboost.dir\src\common\Release\".
  Creating directory "C:\Users\Administrator\Desktop\xgboost\build\CMakeFiles\gpuxgboost.dir\src\linear\Release\".
  Creating directory "C:\Users\Administrator\Desktop\xgboost\build\CMakeFiles\gpuxgboost.dir\src\objective\Release\".
  Creating directory "C:\Users\Administrator\Desktop\xgboost\build\CMakeFiles\gpuxgboost.dir\src\predictor\Release\".
  Creating directory "C:\Users\Administrator\Desktop\xgboost\build\CMakeFiles\gpuxgboost.dir\src\tree\Release\".
CustomBuild:
  Building NVCC (Device) object CMakeFiles/gpuxgboost.dir/src/common/Release/gpuxgboost_generated_host_device_vector.cu
  .obj
  CMake Error at gpuxgboost_generated_host_device_vector.cu.obj.Release.cmake:222 (message):
    Error generating
    C:/Users/Administrator/Desktop/xgboost/build/CMakeFiles/gpuxgboost.dir/src/common/Release/gpuxgboost_generated_host
  _device_vector.cu.obj


C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\VC\VCTargets\Microsoft.CppCommon.targets(171,
5): error MSB6006: "cmd.exe" exited with code 1. [C:\Users\Administrator\Desktop\xgboost\build\gpuxgboost.vcxproj]
Done Building Project "C:\Users\Administrator\Desktop\xgboost\build\gpuxgboost.vcxproj" (default targets) -- FAILED.

Anyone having similar problems?

#2

I got it to work by using CUDA 8.0 + MSVC v140 compilers:

cmake .. -G"Visual Studio 15 2017 Win64" -T v140,cuda=8.0 -DR_LIB=ON -DUSE_CUDA=ON

@RAMitchell It appears that CUDA 9.2 and Visual Studio 2017 may prove problematic for compiling GPU code. Would it be better to update the installation doc?

#3

Yes it would be good to add any notes to the docs re. setups that are know to work. I have installed it with R on windows a total of 1 time so can’t offer much insight.

#4

@hcho3
the same issue happen on mxnet windows version, which also working very well at CUDA 8.0 but fail at CUDA9.2

#5

@hcho3

following official tutorial https://xgboost.readthedocs.io/en/latest/build.html#installing-r-package-with-gpu-support

to build xgboost gpu R-package

cmake .. -DUSE_CUDA=ON -DUSE_NCCL=ON -DR_LIB=ON

error:

-- Found CUDA: /usr/local/cuda (found suitable version "9.0", minimum required is "8.0")
CMake Error at /usr/local/share/cmake-3.6/Modules/FindPackageHandleStandardArgs.cmake:148 (message):
  Could NOT find Nccl (missing: NCCL_INCLUDE_DIR NCCL_LIBRARY)
Call Stack (most recent call first):
  /usr/local/share/cmake-3.6/Modules/FindPackageHandleStandardArgs.cmake:388 (_FPHSA_FAILURE_MESSAGE)
  cmake/modules/FindNccl.cmake:51 (find_package_handle_standard_args)
  CMakeLists.txt:135 (find_package)

I am installing xgboost-gpu on centos7, which seems does not support cuda-90 too. Is any plan to support cuda-90?

#6

You should either install NCCL2 or set -DUSE_NCCL=OFF

1 Like
#7

@hcho3
you are right, it is not CUDA problem, I turned off NCCL, it works. Thanks a lot.

#8

Hi I am facing an issue. I had posted a question on Stack overflow: https://stackoverflow.com/questions/52307360/how-to-compile-xgboost-with-gpu-support-on-windows-10-for-r

#9

I was able to successfully update the libraries in C:\Program Files\R\R-3.5.1\library with GPU support. But when I run tree_method in xgb.cv or xgb.train. Rstudio session crashes.

#10

@rohanadagouda

please directly post more information in here.