R - 3.6.0¶
Basic information¶
- Deploy date: 28 June 2019
- Official Website: https://www.r-project.org/
- License: https://www.r-project.org/Licenses/
- Installed on: Apolo II, Cronos
Dependencies¶
- zlib >= 1.2.5
- bzip2 >= 1.0.6 compiled with
-fPICoption - xz (liblzma) REF https://tukaani.org/xz/
- PCRE >= 8.20 > 10.0 with UTF-8 support
- curl >= 7.22 with https support
- libicu-4.2.1-14.el6.x86_64 and libicu-devel-4.2.1-14.el6.x86_64
- BLAS and LAPACK (Optional) with OpenBlas or MKL
- JDK >= 1.7.0
Installation¶
After the dependencies have been resolved and configured, the R installation can start.
Note
The Intel Compiler will be used to build R.
- Download the selected version from an official mirror (https://cran.r-project.org/mirrors.html).
$ wget https://cran.r-project.org/src/base/R-3/R-3.6.0.tar.gz
$ tar -zxvf R-3.6.0.tar.gz
- Load the corresponding modules to set the building environment.
Note
For BLAS and LAPACK is recommended to use Intel MKL (https://software.intel.com/en-us/articles/build-r-301-with-intel-c-compiler-and-intel-mkl-on-linux)
$ module purge # Clean predefined environment
$ module load zlib bzip2 xz pcre curl java mkl intel
- Configure and build the sources.
$ ./configure --prefix=/share/apps/r/3.6.0/intel-18.0.2/ --build=x86_64-redhat-linux --enable-R-shlib \
--with-blas="-lmkl_rt -liomp5 -lpthread" --with-lapack --enable-BLAS-shlib --without-x --enable-memory-profiling \
LDFLAGS="$LDFLAGS -qopenmp -L/share/apps/bzip2/1.0.6/gcc-5.5.0/lib/ -L/share/apps/pcre/8.41/gcc-5.5.0/lib/ \
-L/share/apps/xz/5.2.3/gcc-5.5.0/lib/" CFLAGS="$CFLAGS -fPIC -qopenmp -O3 -ipo -xHost \
-I/share/apps/bzip2/1.0.6/gcc-5.5.0/include/ -I/share/apps/pcre/8.41/gcc-5.5.0/include/ \
-I/share/apps/xz/5.2.3/gcc-5.5.0/include" CXXFLAGS="$CXXFLAGS -fPIC -qopenmp -O3 -ipo -xHost \
-I/share/apps/bzip2/1.0.6/gcc-5.5.0/include/ -I/share/apps/xz/5.2.3/gcc-5.5.0/include" \
FFLAGS="$FFLAGS -fPIC -qopenmp -O3 -ipo -xHost" FCFLAGS="$FCFLAGS -fPIC -qopenmp -O3 -ipo -xHost"
$ make -j10 && make install
Module¶
The following is the module used for this version.
#%Module1.0#####################################################################
##
## modules r/3.6.0_intel-18.0.2_mkl
##
## /share/apps/modules/r/3.6.0_intel-18.0.2_mkl Written by Johan Yepes
##
proc ModulesHelp { } {
puts stderr "\tR/3.6.0_intel-18.0.2_mkl - sets the Environment for R in \
\n\tthe share directory /share/apps/r/3.6.0/intel_mkl/2017_update-1\n"
}
module-whatis "\n\n\tSets the environment for R language \
\n\tbuilded with Intel Parallel Studio XE Cluster Edition 2018\
\n\t(Update-1) and Intel MKL 2018 (Update-2) version\n"
# for Tcl script use only
set topdir /share/apps/r/3.6.0/intel-18.0.2
set version 3.6.0
set sys x86_64-redhat-linux
conflict r
module load java/jdk-8_u152 mkl/18.0.2 intel/18.0.2
prepend-path PATH $topdir/bin
prepend-path LD_LIBRARY_PATH $topdir/lib64/R/lib
prepend-path LD_RUN_PATH $topdir/lib64/R/lib
prepend-path LIBRARY_PATH $topdir/lib64/R/lib
prepend-path LD_LIBRARY_PATH $topdir/lib64/R/modules
prepend-path LD_RUN_PATH $topdir/lib64/R/modules
prepend-path LIBRARY_PATH $topdir/lib64/R/modules
prepend-path C_INCLUDE_PATH $topdir/lib64/R/include
prepend-path CXX_INCLUDE_PATH $topdir/lib64/R/include
prepend-path CPLUS_INCLUDE_PATH $topdir/lib64/R/include
prepend-path C_INCLUDE_PATH $topdir/lib64/R/include/R_ext
prepend-path CXX_INCLUDE_PATH $topdir/lib64/R/include/R_ext
prepend-path CPLUS_INCLUDE_PATH $topdir/lib64/R/include/R_ext
prepend-path PKG_CONFIG_PATH $topdir/lib64/pkgconfig
prepend-path MAN_PATH $topdir/share/man
Additional Libraries¶
Keras GPU¶
Follow the steps below to set up Keras with GPU support in R.
Note
For the installation, R will be downloaded from Conda because of the Keras library when it is installed does NOT recognize the TensorFlow installation if the R built by us is used.
- Load the Python 3 module.
$ module load python/3.6.8_intel-2019_update-4
- Create the environment in Conda.
$ conda create -n r-tensorflow keras-gpu cudatoolkit=9.2
Warning
It is mandatory to specify the CUDA Toolkit version because the current driver version supports up to CUDA 9.2. If the version is omitted conda will install the latest (CUDA 10.0).
- Activate the environment and install R
$ source activate r-tensorflow
$ conda install r-essentials r-base
- Log in to the GPU node through Slurm.
$ srun -n 1 -t 60 -p accel --gres gpu:4 --pty bash
Warning
You only can log in to the GPU node if you have permissions to launch jobs in the “accel” partition. Ask the administrator if you are not sure.
- Install the Keras library.
$ source activate r-tensorflow
$ R
> install.packages("devtools")
> library(devtools)
> install_github("rstudio/keras")
- Test if GPUs are recognized.
> library(keras)
> k=backend()
> sess = k$get_session()
> sess$list_devices()
- Check if the processes are running in the GPUs.
Note
Login to the GPU node using another terminal session. ssh compute-0-5.
$ watch nvidia-smi
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 396.26 Driver Version: 396.26 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla K80 On | 00000000:05:00.0 Off | 0 |
| N/A 30C P0 73W / 149W | 72MiB / 11441MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 Tesla K80 On | 00000000:06:00.0 Off | 0 |
| N/A 30C P0 73W / 149W | 72MiB / 11441MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 2 Tesla K80 On | 00000000:84:00.0 Off | 0 |
| N/A 34C P0 57W / 149W | 0MiB / 11441MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 3 Tesla K80 On | 00000000:85:00.0 Off | 0 |
| N/A 26C P8 31W / 149W | 0MiB / 11441MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 31882 C ...onda/envs/r-tensorflow/lib/R/bin/exec/R 61MiB |
| 1 31882 C ...onda/envs/r-tensorflow/lib/R/bin/exec/R 61MiB |
+-----------------------------------------------------------------------------+
Authors¶
- Johan Sebastián Yepes Ríos <jyepesr1@eafit.edu.co>