Users are now able to create a Virtual Machine (VM) with GPU support in the Lake Effect cloud. These virtual graphics processing units (vGPU) will provide graphics on a virtual machine's host server rather than on a physical endpoint device


The process of creating this vGPU instance is very similar to that of creating any instance on Lake Effect. If you are unsure about how to create an instance or need more information on our cloud, please view this article


There are a few steps to successfully creating an instance.  First, you should have a key pair file so that you can remotely connect to your instance.  For more information on how to create a key pair file please view this document.  Second, we recommend that our users create a new security group that allows outbound and inbound traffic to/from the instance.  To learn how to do this, please view this document.


Once that's done, we can launch our instance either through the Instances menu or the Images menu. 


From the Instances menu, click on "Launch Instance"



In Details, you will be prompted to choose a name for your instance and change the count if desired. The count is the number of instances you want to create, in most cases this should be left to one. 



Source:

On the Source tab, there are many things you need to consider when deciding how you want your instance created.  

Select Boot Source:  Options are Image, Instance Snapshot, Volume, Volume Snapshot

Create a new volume:  Yes/No (default is Yes which is what we recommend)

Volume Size: This will differ depending on the image selected. If you leave it at default then it will create an image large enough for the root volume and some extra space. The larger the volume size you select, the longer it will take for your instance to start up. We recommend leaving it at default and attaching additional volumes if you need a data directory. Each project does have a storage quota. See your usage and available space in the Dashboard view.

Delete Volume on Instance Delete:  Default is No (we recommend selecting Yes unless you may want to use this volume attached to other instances.  Keep in mind, the volume will be left in your account, counting against your storage quota, when your instance is deleted.)

Device name:  Default is vda (linux volume name)


You will also be prompted to pick an image that you want to run.  We recommend that you choose "CentOS-7.7.1908-GPU-CUDA-10.2". This image has the Nvidia vGRID driver version 440.56 and Cuda 10.2 built into it.  It also has the client license utility installed that automatically checks out an Nvidia vGrid license. The GPU will still work without a license, however the full feature set is not available, performance is degraded and you will get license warnings when you try and use it.   If, for some reason you require a different image, you will have to do the necessary configurations and driver installations to make the image compatible with the GPU hardware.  It is critical that the version of the driver that runs on your instance is the same major version that is installed on CCR's cloud infrastructure that hosts the cloud instances, otherwise you will have incompatibilities and it will not work correctly.  Documentation is not provided for this; however, CCR staff are available for consultation.  Please see our rates and policies here. 


Once that's done, click on the arrow corresponding to the image to add.


Flavor:

This section displays all of the available options for sizing your instance.  You must use the up arrow next to the instance "flavor" you decide you want to run.  This list shows each flavor, the number of virtual CPUs (VCPUS), RAM (memory), total disk size, and total size of root disk.  If you click the arrow to the left of the flavor name, you will be shown how starting up an instance of that size will impact your group's (project) quota. For the sake of creating a vGPU, you can either choose the g1.v100d-2q or g1.v100d-4q flavor (where 2q and 4q mean 2 GB and 4GB of GPU memory respectively). 


Networks:


There are two types of network on Lake Effect: public, and private. The default is the lakeeffect-199.109.195 public network, but if you want to create a private network for your instance/project and assign it a floating IP address, please view this document on how to do so.


Network Ports:


At this time there are no virtual switch ports configured in CCR's LakeEffect cloud so this step can be skipped.  We are currently testing this functionality of Openstack.  Please contact CCR help if this is something you'd be interested in utilizing.

Security Groups:


Security groups are an important concept to understand in Openstack.  Security groups define a set of IP filter rules that determine how network traffic flows to and from an instance, like a firewall.  As explained earlier in this article, it is important that you create a new security group that allows both outbound and inbound traffic through instance because the default rule in LakeEffect only allows outbound traffic - no inbound traffic is permitted.  If the default security group is associated with an instance you will have no way to connect to it after it launches. 


Once you have created your new security group, you can pick it by clicking on the arrow next to it (in our case we went with "test"- a newly created security group).


Key Pair:


A key pair allows you to SSH into your instance.  You may select an existing key pair, import a key pair you've already created (see how-to article on creating key pairs), or generate a new key pair from this tab.



Configuration:


This tab allows you to script certain configuration details when an instance is launched.  Please click the question mark icon on this tab for more details.  This is for advanced users only.


Some users may want to partition the root disk of an instance in a specific way.  To do this, change the Automatic to Manual under Disk Partition and you will be prompted to setup the partitions when the instance is launched.


The remaining three tabs (server groups, scheduler hints, and metadata) on the launch instance GUI are not necessary. You can now click on "Launch Instance" to get your virtual machine running. 


Note: Your instance might take a while (anywhere from few seconds to a minute) to get started


Once the Power status of your instance is "Running", you will be able to SSH into your instance. In order to do that, just navigate to the directory you saved your key pair file to and run this command:


ssh -i <key_name.pem> <username>@<instance IP address>

See this article for default usernames on new instances


example:

ssh -i cloud.pem ubuntu@199.109.192.2



For the sake of this article, we created a key pair called "gputesting", a security group called "test", and chose image CentOS-7.7.1908-GPU-CUDA-10.2. Hence, our command would look something like this:


ssh -i gputesting.pem centos@199.109.192.2 


To test or to look up the info on your GPU, please use this command: 

nvidia-smi