研究ノート GPU CPU Central Processing Unit CPU BTO Build To Order CPU CPU GPGPU Tesla GPU Graphics Processing Units GPGPU General-Purpose computing on Grap

Size: px

Start display at page:

Download "研究ノート GPU CPU Central Processing Unit CPU BTO Build To Order CPU CPU GPGPU Tesla GPU Graphics Processing Units GPGPU General-Purpose computing on Grap"

ありさおおばま
5 years ago
Views:

1 1 7 GPU Getting Started with GPGPU

2 研究ノート GPU CPU Central Processing Unit CPU BTO Build To Order CPU CPU GPGPU Tesla GPU Graphics Processing Units GPGPU General-Purpose computing on Graphics Processing Units GPU GPGPU, CUDA, Python, Ubuntu, Docker 目次 Python GPGPU OpenSSH Winodows GPGPU Ubuntu Docker LXD Docker 1 はじめに CPU CPU

3 Vol. No. BTO CPU CPU GPGPU Tesla GPU GPGPU GPU GPGPU GPGPU GPU GPU AMD NVIDIA NVIDIA GeForce G Ubuntu. GPGPU PC Windows Home G G Windows Python Ubuntu. Windows Ubuntu. Ubuntu. Windows Del UEFI UEFI Ubuntu. LTS Ubuntu. LTS Remix iso GeForce Intel X CPU LGA - CPU GeForce Ubuntu. LTS Remix iso Live USB Ubuntu. Ubuntu. Remix iso Live USB Ubuntu. Windows HOME Ubuntu VirtualBox Virtual Ubuntu Network Virtual PC Windows Anniversary Update WSL Windows Subystem for Linux WSL Ubuntu WSL Ubuntu CLI X Windows System GUI FireFox Virtual WSL GPGPU Mac Linux Windows Microsoft Linux Foundation Linux

4 GPU GPGPU NVIDIA CUDA Ubuntu NVIDIA GPU nouveau GPGPU NVIDIA Linux GPU CUDA UEFI nouveau DKMS Ubuntu GeForce CUDA Toolkit GPU nvidia-smi GPU ID, GeForce GTX..., C PCI Bus ID : :. MiB CUDA Tookit CUDA CUDA /usr/local/ cuda-. /NVIDIA CUDA-. Samples Utilities/deviceQuery CUDA OpenCL CUDA GUI Ctrl+Alt+F GUI CLI NVIDIA purge Windows Grub Windows Windows LiveUSB

5 Vol. No. sudo make devicequery./devicequery 2 Python で GPGPU CUDA GPGPU Python GPGPU Python Continum Analytics Anaconda Python CUDA gcc. gcc downgrade gcc-. g++-., /usr/local/ cuda/ bin/gcc / usr / local /cuda / bin / g++ make Anaconda Download for Linux Python. version -BIT INSTALLER MD or SHA- terminal bash Anaconda -.. -Linux-x.sh Tab terminal python Python Anaconda Anaconda package manager environment manager conda conda conda update conda update

6 GPU PyCUDA CuPy Numpy GPU Chainer Deep Learning Python GPU Chainer, Chainer python -c import chainer conda conda search chainer conda conda pip pip install chainer MNIST example python train mnist.py -g= gpu CPU CPU GPU GPU 3 OpenSSH で Windows ノートから GPGPU GPGPU Windows GPU Windows GPGPU GPU PC Windows PC PC OpenSSH GPU PC Ubuntu SSH Windows Windows OpenSSH SSH Ubuntu PC SSH SSH Windows Anaconda Python. NVIDIA cudnn Deep Neural Network library enable conda environment management Chainer Python.. import cupy.cudnn cudnn enable cudnn NVIDIA Accelerated Computing Program

7 Vol. No.. Ubuntu PC SSH sudovapt-getvinstallvssh PC ip ifconfig. Windows OpenSSH OpenSSH bit OpenSSH-Win.zip Path. Windows ssh-keygen.\ssh-keygen.exe public/private rsa key Enter passphrase passphrase login id - rsa id rsa.pub. id - rsa.pub Windows Ubuntu PC scp scpvid_rsa.pubvserver-ip-address:~/. Ubuntu PC id_rsa.pub authorized.keys sudo root sudovmkdirv.ssh sudovcatvid_rsa.pubv>>v.ssh/authorized.keys Windows SSH GPGPU. Windows sshvuser@host-ip-adress passphrase passphrase user Ubuntu PC user password Ubuntu PC UTF- Windows exit

8 GPU chcp UTF- 4 Ubuntu の Docker と LXD を利用する Mac Mac OSX OS OS OS USB HDD PC macos Sierra SSD MAMP moodle STACK G SSD G MAMP moodle STACK, Maxima, Gnuplot, Xterm OS Virtual macosx Ubuntu Windows OS OS OS, OS Mac OSX OS VirtualBox Windows Ubuntu OS OS OS OS OS OS Docker Docker LXD LXC API OS Ubuntu Ubuntu

9 Vol. No. Ubuntu NVIDIA NVIDIA CUDA cudnn Chainer Tensorflow, Tensorflow nvidia-docker Tensorflow Github nvidia-docker Installation The list of prerequisites for running nvidia-docker is described below.. GNU/Linux x with kernel version > :. Docker>= : official docker-engine only. NVIDIA GPU with Architecture> Fermi.. NVIDIA drivers >= : with binary nvidia-modprobe Your driver version might limit your CUDA capabilities see CUDA requirements CUDA toolkit version. Driver version. kernel uname -rv. NVIDIA GPU GeForce G Pascal Architecture NVIDIA nvidia-smi. nvidia-docker nvidia-docker is a thin wrapper on top of docker and act as a drop-in replacement for the docker command line interface. This binary is provided as a convenience to automatically detect and setup GPU containers leveraging NVIDIA hardware. Internally, nvidia-docker calls docker and relies on the NVIDIA Docker plugin to discover driver files and GPU devices. Note that nvidia-docker only modifies the behavior of the run and create Docker commands. All the other commands are just pass-through to the docker command line interface. As a result, you can t execute GPU code when building a Docker image.if the nvidia-docker-plugin is installed on your host and running locally, no additional step is needed. nvidia-docker will perform what is necessary by querying the plugin when containers using NVIDIA GPUs need to be launched. nvidia-docker docker command line interface docker run create nvidia-docker GPU nvidia-smi GPU nvidia-docker-plugin NVIDIA architecture Tesla,Fermi,Kepler,Maxwell Pascal Volta. Compute Capability Pascal archiitecture GeForce GTX Compute Capability.

10 GPU Docker. official docker-engine only official docker-engine Ubuntu docker.io Docker docker-engine Docker Docker CommunityEdition Docker CE nvidia-docker Quick start dpkg -i stvrc file-rc make install sudovdockervimages nvidia-docker build nvidia-docker GPU nvidia-smi sudovnvidia-dockervrunvnvidia/cudavnvidia-smi plugin not found sudovnvidia-docker-plugin sudovnvidia-dockervrunvnvidia/cudavnvidia-smi GPU. GPU docker-ce nvidia-docker nvidia-docker TensorFlow Tensoflow github Docker Docker Hub tensorflow gcr.io/tensorflow/tensorflow:xxx-gpu TensorFlow with all dependecies and support for NVidia CUDA gpu gpu CPU GPU xxx sudovnvidia-dockervrunv-itdv--name=tensorflow-xxx-gpuv-pv8888:8888v -pv6006:6006vgcr.io/tensorflow/tensorflow:xxx-gpu sudovdockervimages sudovdockervpsv-a -a docker gcr.io/tensorflow/tensorflow : xxx-gpu tensorflow-xxx-gpu docker docker

Vol. No. sudovnvidia-dockervexecv-itvtensorflow-xxx-gpuvbash docker bash -it $ # root@docker-id:/notebooks# Github tensorflow git git gitvclonev-bvr0.

11 Vol. No. sudovnvidia-dockervexecv-itvtensorflow-xxx-gpuvbash docker bash -it $ # root@docker-id:/notebooks# Github tensorflow git git gitvclonev-bvr0.10v--single-branchv--recurse-submodulesvhttps:// git.com/tensorflow/tensorflow.git tensorflow ls cd tensorflow/tensorflow/models/image/cifar10/ Tab pythonvcifar10_multi_gpu_train.pyv--num_gpus=1v--vmax_steps=1000 cuda sudo nvidia-docker-plugin sudo docker start tensorflow-xxx-gpu docker sudo nvidia-docker exec docker #!/bin/bash sudo nvidia-docker-plugin & & sudo docker start tensorflow-xxx-gpu sudo docker exec -it tensorflow-xxx-gpu bash bash nvidia-docker.sh tensorflow-xxx-gpu docker notebooks

GPU GPU nvidia-smi GPU W GPU MiB MiB python Process MiB Volatile GPU- Util % GeForce GTX G GeForce GTX G tensorflow-xxx-gpu jupyter notebook Firefox Firefox http://localhost: jupyter notebook Home

12 GPU GPU nvidia-smi GPU W GPU MiB MiB python Process MiB Volatile GPU- Util % GeForce GTX G GeForce GTX G tensorflow-xxx-gpu jupyter notebook Firefox Firefox jupyter notebook Home hello tenssorflow. ipynb mnist from scratch 4.1 TensorBoard を使う tensorflow TensorBoard TensorBoard TensorFlow docker tensorboard Starting TensorBoard on port You can navigate to : docker firefox URL... :

13 Vol. No. TensorBoard 5 Docker とはどんなもの Docker Docker PC works on my machine VM VM OS Docker nvidia-docker Tensorflow GPU Python Docker Docker Docker image container sevice stack swarm Docker Hub Registry Image Dockerfile mkdir Docker Hub Image username/repository:tag docker run Image name= -p mapping Web Web Browser Get started with Docker Part :Orientation Docker. IP addresses and Ports. Virtual Machines

14 GPU. Editing configuration files. Basic familiarity with the ideas of code dependencies and building. Machine resource usage terms, like CPU percentages, RAM use in bytes,etc. terminal Linux cd, ls,cp,mkdir,cat,curl. vi Get started Part : Containers friendyhello Docker Docker public registry registry $sudovdockervrunv-dv-pv5000:5000v-vv/var/opt:/var/lib/registryvregistry:2.3.0 registry public registry registry.. -d -p -v /var/opt / var/lib/registry /var/opt friendlyhello registry localhost: repository $sudovdockervtagvfriendlyhellovlocalhost:5000/repository /friendlyhello $sudovdockervimages localhost: /repository /friendlyhello push $sudovdockervpushvlocalhost:5000/repository /friendlyhello /var/opt docker/registry/v / repositoires/repositry /friendlyhello registry $sudovdockervpullvlocalhost:5000/repository /friendlyhello Get Started, Part :Swarms 5.1 Docker for Windows について Docker for Windows Windows

15 Vol. No. Microsoft bit windows pro x Hyper-V VirtualBox Windows Hyer-V Windows Hyper-V VM VirtualBox Ubuntu. LTS Docker Docker for Windows Windows InstallDocker.msi cmd Docker docker version docker.. -ce OS/Arch Client windows/amd Server linux/amd Hyper-V MobyLinuxVM MB Docker forwindows cmd docker run hello-world Hello from Docker! This message shows that your installation appears to be working correctly. docker Windows docker 6 おわりに Python GPGPU GPGPU Linux Ubuntu OS NVIDIA GPU CUDA,Python Linux Docker Deployment GeForce GTX G GPU GPU CPU GPU. CPU GPU NVIDIA GeForce GTX G W W GeForce GTX Ti G W W GPU PC W

16 GPU GPU Windows kernel with the Anniversary Update WSL Windows Subsystem for Linux Ubuntu Hyper-V Linux Docker Windows containers Linux Windows Windows kernel with the Creators Update WSL Ubuntu. Linux Hyper-V Windows Linux Hyper-V Windows Pro, Enterprise,Educaton Windows Home Pro Windows Windows Pro PC

28 Docker Design and Implementation of Program Evaluation System Using Docker Virtualized Environment

28 Docker Design and Implementation of Program Evaluation System Using Docker Virtualized Environment 1170288 2017 2 28 Docker,.,,.,,.,,.,. Docker.,..,., Web, Web.,.,.,, CPU,,. i ., OS..,, OS, VirtualBox,.,