Parallel Random Number Generation on GPU.
This clrng R package is built as a tool set for parallel random number generation on GPUs. It is dependent on the gpuR package and utilize the MRG31k3p RNG from the clRNG, an OpenCL library by Pierre L’Ecuyer.
Installing drivers
The clrng and gpuR packages require GPU drivers and opencl. This can be tricky. Although the package is only officially supported on unix, it should work on any computer where openCL is available.
Check you have a GPU
lspci -kv
Below are some instructions for installing the required drivers for Ubuntu.
Nvidia
See the Nvidia installation guide
If you're lucky, the following will work.
sudo apt install nvidia-driver-535
sudo reboot
sudo apt-get install cuda
Check you've installed the drivers correctly.
nvidia-smi
Install openCL
sudo apt install -y nvidia-opencl-dev clinfo
check openCL is working
clinfo
AMD
See the AMD rocm install guide
sudo apt install amdgpu-dkms
sudo apt install rocm
sudo reboot
Check the drivers
/opt/rocm/bin/rocm-smi
... and openCL
/opt/rocm/bin/clinfo
Amazon AWS
Start an instance with
g3s.xlarge- ubuntu server 22.04
- create security group, allow ssh traffic from anywhere, https from anywhere
- 20gb storage
Make sure the software is up to date
sudo apt update
sudo apt dist-upgrade
sudo reboot
Install a version of the kernel with the DRM module
sudo apt install -y linux-headers-virtual linux-source linux-image-extra-virtual
sudo apt autoremove
sudo apt clean
sudo reboot
Set up the nvidia software repository
wget -O /tmp/cuda-keyring.deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i /tmp/cuda-keyring.deb
sudo apt update
Install cuda
sudo apt install -y cuda
sudo apt autoremove
sudo apt clean
sudo reboot
Check the driver is working
nvidia-smi
Install openCL
sudo apt install -y nvidia-opencl-dev clinfo
Check openCL is working
clinfo
Digital Research Alliance of Canada cloud
Free for academics in Canada: info
Start an instance with
g1-8gb-c4-22gb- Ubuntu
Update and reboot
sudo apt update
sudo apt dist-upgrade
sudo reboot
Add the repository for GPU drivers
wget -O /tmp/arbutus-cloud-repo_all.deb http://repo.arbutus.cloud.computecanada.ca/pulp/deb/ubuntu22/pool/main/arbutus-cloud-repo_0.2_all.deb
sudo dpkg -i /tmp/arbutus-cloud-repo_all.deb
Install drivers
sudo apt --yes install nvidia-vgpu-kmod nvidia-vgpu-tools nvidia-vgpu-gridd
sudo reboot
Check the driver is working
nvidia-smi
Install openCL
sudo apt-get install --yes opencl-headers clinfo ocl-icd-opencl-dev
Check openCL is working
clinfo
R and packages
Add R repositories
sudo add-apt-repository --yes "deb https://cloud.r-project.org/bin/linux/ubuntu $(lsb_release -cs)-cran40/"
wget -qO- https://cloud.r-project.org/bin/linux/ubuntu/marutter_pubkey.asc | sudo tee -a /etc/apt/trusted.gpg.d/cran_ubuntu_key.asc
sudo add-apt-repository --yes ppa:c2d4u.team/c2d4u4.0+
Install R
sudo apt --yes install r-cran-devtools r-cran-rcppeigen r-cran-bh r-cran-testthat r-cran-knitr r-cran-assertive
sudo apt-get clean
Set up some folders for R, including a personal library.
mkdir ~/.R
echo 'MAKEFLAGS = -j4' > ~/.R/Makevars
R -e 'dir.create(Sys.getenv("R_LIBS_USER"), recursive = TRUE)'
Install the GPU packages!
devtools::install_github("eborgnine/gpuR")
devtools::install_github("ruoyongxu/clrng")
openCL on CPU
It is possible to install openCL for use with a CPU rather than a GPU. This could be useful for development and testing, but the code will run considerably slower than on a GPU.
sudo apt install intel-opencl-icd
sudo apt install opencl-dev clinfo
clinfo
library('gpuR')
listContexts()
setContext(grep("cpu", listContexts()$device_type)[1])