gce-gpu

create instance & firewall

• new

gcloud deployment-manager deployments create gpu-instance --template ./deployment/vm.jinja --properties name:vm01,zone:us-central1-b,instanceType:custom-2-4096,diskSizeGb:50,gpuCount:1
gcloud deployment-manager deployments create firewall --template ./deployment/firewall.jinja --properties sourceIp:[アクセス元IPアドレス]

• update

gcloud deployment-manager deployments update gpu-instance --template ./deployment/vm.jinja --properties name:vm01,zone:us-central1-b,instanceType:custom-2-4096,diskSizeGb:50,gpuCount:1
gcloud deployment-manager deployments update firewall --template ./deployment/firewall.jinja --properties sourceIp:[アクセス元IPアドレス]

git clone

gcloud compute ssh vm01 --zone=us-central1-b --command="git clone https://github.com/bassbone/gce-gpu.git"

setup nvidia

gcloud compute ssh vm01 --zone=us-central1-b --command="sh gce-gpu/setup/nvidia-install.sh"

reset instance

gcloud compute instances reset vm01 --zone=us-central1-b
sleep 60 # for wait reboot.

setup docker

gcloud compute ssh vm01 --zone=us-central1-b --command="sh gce-gpu/setup/docker-install.sh"

docker build & up

gcloud compute ssh vm01 --zone=us-central1-b --command="cd gce-gpu; make start"

(git update)

gcloud compute ssh vm01 --zone=us-central1-b --command="cd gce-gpu; git pull origin master"