This tutorial shows how to create a virtual Linux workstation running CentOS with an attached display-capable GPU. (To create a Windows Workstation, see the tutorial for Creating a virtual GPU accelerated Windows workstation.) Google Cloud offers three display-capable GPUs: NVIDIA T4, NVIDIA Tesla P4, and NVIDIA Tesla P100.
After you create the virtual workstation, you learn how to remotely access it using Teradici PC-over-IP (PCoIP), a remote desktop protocol widely used in the media and entertainment industry. PCoIP offers features essential to media production workloads, such as color accuracy and support for lossless display.
- Create a Compute Engine instance with a GPU. This instance serves as the foundation for a virtual workstation.
- Install NVIDIA drivers on the virtual workstation.
- Install Teradici Cloud Access Software on the virtual workstation.
- Connect to the virtual workstation from your Mac or Windows computer using a PCoIP software client or Zero Client, a type of hardware endpoint. To access the virtual workstation from a Linux computer, contact your Teradici representative.
This tutorial uses the following billable components of Google Cloud:
You can use the pricing calculator to generate a cost estimate based on your projected usage. As of the time of writing, the approximate cost for a typical media workstation configuration illustrated in this tutorial is US$1.36 per hour.
The resources that make up the virtual workstation and the factors that affect cost in this tutorial are:
- 24 vCPUs, 32 GB RAM custom machine type.
- 100-GB SSD persistent boot disk.
- NVIDIA Tesla P4 Virtual Workstation GPU.
- Internet egress.
Internet egress represents data that streams from your virtual workstation to your local display client and is billed at internet egress rates. Variables that affect data egress during a PCoIP session are bandwidth, screen resolution, number of display monitors, applications used, and the type of activity on each monitor. The cost in the example is based on an average usage of 10 Mbps. Teradici's Workflow Planning Guide can help you understand different workload requirements.
Before you begin
This tutorial uses
gsutil commands, which you can run from a
instance launched from the
If you want to use
gsutil on your local workstation, install the
The tutorial shows you how to run commands in Cloud Shell; if you
use the Cloud SDK on your workstation, adjust the instructions
Sign in to your Google Account.
If you don't already have one, sign up for a new account.
In the Google Cloud Console, on the project selector page, select or create a Google Cloud project.
Make sure that billing is enabled for your Cloud project. Learn how to confirm that billing is enabled for your project.
- Enable the Compute Engine API.
In addition, make sure you have the following:
- A Google Cloud project with quota for
virtual workstation GPUs
in your selected
You can get a listing of GPU availability using the
gcloud compute accelerator-types listcommand.
- A Google Chrome browser to access the Cloud Console.
- A Teradici Zero Client or the latest Teradici software client for Windows, Mac, or Linux to access the virtual workstation.
- A Teradici Cloud Access Software license. You can sign up for a trial license, or contact your Teradici representative. You will be provided with a 30-day trial registration code to use for this virtual workstation.
Understanding the architecture
The following diagram shows the components that are used in this tutorial to deploy a single virtual workstation. Optional components shown in the diagram include different ways to connect to your virtual workstation, object and shared storage, and an additional instance for serving third-party licenses.
Choosing an accelerator
An important factor when you deploy a virtual workstation is the latency between your location and the instance you create. The lower the latency, the better the experience. Therefore, you want to work in a zone that's geographically closest to you. To learn more about regions and zones, see Geography and regions.
Open Cloud Shell. (If you're using the Cloud SDK, open a terminal window on your computer.)
Get a list of the zones in which GPUs are available:
gcloud compute accelerator-types list
Take note of the zone that's physically closest to you.
Set the zone that you want to work with:
gcloud config set compute/zone zone
Replace zone with the name of the zone you're using, such as
Choosing a machine type
You can attach P4 GPUs to any machine type, but each GPU that's added to a virtual workstation must be coupled with a minimum number of vCPUs. This prevents oversubscription of any one resource. For the NVIDIA Tesla P4, you can attach up to 24 vCPUs to 1 GPU. To add more than 24 vCPUs to your virtual workstation, you must add an additional P4 GPU for every additional 24 vCPUs. For example, a 48-vCPU virtual workstation would require you to attach 2 P4 GPUs.
The example in this tutorial consists of a 24-vCPU virtual workstation, which is the maximum number of vCPUs allowed per NVIDIA Tesla P4 GPU.
Creating the virtual workstation
Teradici Graphics Agent (which you install on your virtual workstation later in this tutorial) requires you to enable IP forwarding and to allow HTTPS server traffic during virtual workstation creation.
In Cloud Shell, create the Compute Engine virtual workstation instance. You must provide values for the placeholders such as
gcloud compute instances create name \ --machine-type machine-type \ --accelerator type=accelerator,count=num-gpus \ --can-ip-forward \ --maintenance-policy "TERMINATE" \ --tags "https-server" \ --image-project centos-cloud \ --image-family centos-7 \ --boot-disk-size size
gcloud compute instances create test-vws \ --machine-type custom-24-32768 \ --accelerator type=nvidia-tesla-p4-vws,count=1 \ --can-ip-forward \ --maintenance-policy "TERMINATE" \ --tags "https-server" \ --image-project centos-cloud \ --image-family centos-7 \ --boot-disk-size 100
After the virtual workstation is created, the machine status is displayed. The output looks similar to the following:
NAME ZONE MACHINE_TYPE PREEMPTIBLE INTERNAL_IP EXTERNAL_IP STATUS test-vws us-west2-b custom (24 vCPU, 32.00 GiB) 10.168.0.3 XX.XXX.XX.XXX RUNNING
The virtual workstation is created in your project's default VPC network. If you want to create your virtual workstation in a different VPC network, add the following flag to the command:
networkwith the name of the network to use.
Note the virtual workstation's external IP address. You will use it later in the tutorial.
Signing in to your virtual workstation
After you create the virtual workstation, you sign in to the machine so that you can configure it.
In Cloud Shell, connect to the new virtual workstation:
gcloud compute ssh test-vws
Set your account password. Teradici PCoIP requires a user password to be set.
sudo passwd `whoami`
When you're prompted, enter a password. You will use this password later in the tutorial to log on to your virtual workstation with the Teradici PCoIP Client.
Installing graphics libraries and a window manager
The default Google Cloud CentOS 7 image is a minimal installation of CentOS 7.x. The next step is to install libraries that are necessary to run your virtual workstation as a graphics workstation. For this tutorial, you also install the KDE window manager.
Install the required components:
sudo yum -y update sudo yum -y install kernel-devel sudo yum -y groupinstall "KDE desktop" "X Window System" "Fonts" sudo yum -y groupinstall "Development Tools" sudo yum -y groupinstall "Server with GUI"
Reboot the workstation:
Your connection from Cloud Shell will be closed.
Installing the NVIDIA driver
NVIDIA T4, NVIDIA Tesla P4, and NVIDIA Tesla P100 GPUs work on Google Cloud only with qualified NVIDIA Quadro Virtual Data Center Workstation (vWS) drivers for both compute and display workloads. These drivers can be downloaded from a public storage bucket.
When the virtual workstation has restarted, in Cloud Shell, reconnect to the virtual workstation:
gcloud compute ssh test-vws
On your virtual workstation, use
gsutilto get a listing of the latest drivers:
gsutil ls gs://nvidia-drivers-us-public/GRID
For this tutorial, you use the latest graphics drivers that are available at the time of writing: GRID11.1 Linux driver (version 450.80.02). The latest qualified driver can always be found under the
GRIDdirectory with the highest version number; if you find a newer driver, use that one.
Download and install the driver. If you're using a version of the driver that's more recent than 450.80.02, change the command accordingly.
curl -O \ https://storage.googleapis.com/nvidia-drivers-us-public/GRID/GRID11.1/NVIDIA-Linux-x86_64-450.80.02-grid.run sudo bash NVIDIA-Linux-x86_64-450.80.02-grid.run
During driver installation, you might see some prompts:
- If you're prompted to install 32-bit binaries, choose Yes.
- If you're prompted to modify the
x.orgfile, choose No.
Verify the driver is installed and working:
The output is similar to the following:
Thu Sep 20 21:58:23 2018 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 450.80.02 Driver Version: 450.80.02 CUDA Version: 11.0 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |===============================+======================+======================| | 0 Tesla P4 On | 00000000:00:04.0 Off | 0 | | N/A 41C P0 23W / 75W | 0MiB / 7611MiB | 0% Default | | | | N/A | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=============================================================================| | No running processes found | +-----------------------------------------------------------------------------+
If you don't see output that's similar to this listing, refer to the Troubleshooting section later in this tutorial.
Installing Teradici Cloud Access Software
Teradici Cloud Access Software provides a graphics agent that runs on your virtual workstation, delivering the desktop to your hardware or software client.
On your virtual workstation, add the Teradici software:
sudo rpm --import https://downloads.teradici.com/rhel/teradici.pub.gpg sudo yum -y install wget sudo wget -O /etc/yum.repos.d/pcoip.repo \ https://downloads.teradici.com/rhel/pcoip.repo
Update the software repositories:
sudo yum -y update
Install the Teradici Cloud Access Software:
sudo yum -y install pcoip-agent-graphics
Set display state to
sudo systemctl set-default graphical.target
Reboot the virtual workstation:
Register Teradici Graphics Agent
To use Teradici Graphics Agent, you must have a license, as noted earlier in the tutorial.
In Cloud Shell, reconnect to the virtual workstation:
gcloud compute ssh test-vws
On your virtual workstation, activate your Teradici Cloud Access Software license:
Creating a firewall rule
The PCoIP client communicates with your virtual workstation using several ports. You must set firewall rules that allow traffic to and from your virtual workstation.
In Cloud Shell, create a firewall rule that opens the required ports:
gcloud compute firewall-rules create allow-teradici \ --allow tcp:443,tcp:4172,udp:4172,tcp:60443
Signing in to your virtual workstation using the PCoIP client
On your local computer, go to the PCoIP Clients section on the Teradici support page, and then download, install, and launch the PCoIP Client application for your operating system.
Select New Connection.
In the Host Address field, enter the external IP address of your virtual workstation. If you want, you can enter a name for the connection.
When you are connected, authenticate by entering the username and password that you created earlier for the virtual workstation.
Select the desktop to run and then click Connect.
In a few seconds, you see your Linux desktop.
Testing your virtual workstation
After you've deployed your virtual workstation, you can test performance and interactivity using a number of tools:
- Run GPU benchmark tools, such as glxgears or glmark2, which are simple programs to test graphics performance on a Linux workstation.
- Install Blender, an open source 3D software package.
- Run render benchmarking tools for popular renderers such as V-Ray, Octane, or Maxon.
- Use Google Chrome to browse your favorite sites or play YouTube videos.
This section lists issues you might encounter when you set up or connect to the workstation.
NVIDIA-SMI has failed
Issue: NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver.
Solution: Reinstall the driver and try running
nvidia-smi again. If the
command still fails, try uninstalling the NVIDIA driver, installing the
module, and then reinstalling the driver. Doing this registers the dkms module
into the kernel so that updates of the kernel won't require a reinstallation
of the graphics driver.
X11 doesn't start
Issue: X11 doesn't start as intended.
Solution: Ensure that the Nouveau graphics driver isn't present in your installation. Nouveau drivers are open source graphics drivers included in some Linux versions. To check if the Nouveau driver is present, run this command in the virtual workstation shell:
lsmod | grep nouveau
If you see any results, follow these steps to disable loading of the Nouveau driver before you install the qualified NVIDIA graphics driver:
As the superuser in the virtual workstation shell, open a text editor and add the following to the last line in the file
Reboot the virtual workstation:
When the virtual workstation is rebooted, reconnect to it from Cloud Shell:
gcloud compute ssh test-vws
As the superuser, edit the file
Find the entry named
GRUB_CMDLINE_LINUXentry, or add it if it's not already there, and then add the following value to the entry:
Make sure that this value is inside the quotation marks. For example:
GRUB_CMDLINE_LINUX="crashkernel=auto console=ttyS0,38400n8 rd.driver.blacklist=nouveau nouveau.modeset=0"
Generate a new grub configuration to include the changes:
sudo grub2-mkconfig -o /boot/grub2/grub.cfg
Reboot your virtual workstation again:
After the virtual workstation has restarted, reconnect to it from Cloud Shell:
gcloud compute ssh test-vws
Ensure that the Nouveau driver is no longer present:
lsmod | grep nouveau
A blank string means that the Nouveau driver is not installed. For more information on the Nouveau driver, see Common Problems in the NVIDIA documentation.
Unable to connect to the virtual workstation
Issue: You are using a PCoIP Zero Client, and you are unable to connect to your virtual workstation.
Solution: Ensure that your Zero Client has firmware version 6.1 or later installed before you connect to the virtual workstation. For more information, contact your Teradici representative.
To avoid incurring charges to your Google Cloud Platform account for the resources used in this tutorial:
After you've finished the tutorial, clean up the resources you created on Google Cloud so you won't be billed for them in the future.
Stop your virtual workstation
Stopped virtual workstations incur costs for persistent disk, but can be restarted at any time. To stop your virtual workstation, run the following command in Cloud Shell:
gcloud compute instances stop test-vws
Delete the project
- In the Cloud Console, go to the Manage resources page.
- In the project list, select the project that you want to delete, and then click Delete.
- In the dialog, type the project ID, and then click Shut down to delete the project.
Delete all the components
- Learn how to create a virtual GPU-accelerated Windows workstation.
- Learn more about NVIDIA GRID GPUs for graphics workloads.
- Learn more about NVIDIA Quadro Virtual Data Center Workstation software.
- Learn more about Teradici Cloud Access Software.
- Learn more about how Teradici PCoIP differs from other remote desktop protocols.
- Try out Teradici's preconfigured virtual workstations for Windows and Linux in Google Cloud Marketplace.
- Try out other Google Cloud features for yourself. Have a look at our tutorials.