In order to fully utilize the Virtual GPUs in LakeEffect, you must use the Nvidia vGRID Driver and register the Instance with the CCR  NVidia License server. If this is not done correctly, you will have an unlicensed vGPU that runs a demo feature set and the performance is significantly degraded.


Perform the following steps after the Instance has been created.


NOTE: You must use one of the flavors that has a Virtual GPU assigned to it. LakeEffect has a few different options for vGPU. The following are vGPU flavors available:


g1.t4-4c: 10 vCPU, 23GB RAM, 50GB Storage, NVidia Tesla T4 with 4GB Memory


g1.t4-8c: 20 vCPU, 46GB RAM, 50GB Storage, NVidia Tesla T4 with 8GB Memory


g1.v100d-4c: 10 vCPU, 23GB RAM, 50GB Storage, NVidia Tesla V100 with 4GB Memory


g1.v100d-8c: 20 vCPU, 46GB RAM, 50GB Storage, NVidia Tesla V100 with 8GB Memory



NOTE: The following example is for an Instance running Ubuntu 20.04



1. Verify that a GPU device has been attached to the Instance (this will depend on what flavor you used):

$ lspci |grep NVIDIA
   00:05.0 3D controller: NVIDIA Corporation TU104GL [Tesla T4] (rev a1)

2. Update to the latest set of updates and reboot:

$ sudo apt -y update
$ sudo apt -y upgrade

3. Log onto instance and install the supporting packages for the Nvidia vGRID Driver

$ sudo apt -y install dkms

4. Download and Install the Nvidia vGRID driver from CCR Repos:

$ wget http://mirrors.ccr.buffalo.edu/ccr/openstack/vgpu/drivers/NVIDIA-Linux-x86_64-470.63.01-grid.run
$ sudo sh ./NVIDIA-Linux-x86_64-470.63.01-grid.run -s --disable-nouveau --dkms --no-install-compat32-libs

5. Download and Install Client Configuration file and Client Token

$ wget http://mirrors.ccr.buffalo.edu/ccr/openstack/vgpu/configs/client_configuration_token.tok
$ wget http://mirrors.ccr.buffalo.edu/ccr/openstack/vgpu/configs/gridd.conf-nvdls
$ sudo mv gridd.conf-nvdls /etc/nvidia/gridd.conf
$ sudo mv client_configuration_token.tok /etc/nvidia/ClientConfigToken/

6. Reboot the VM

$ sudo reboot

7. Check to see if the vGPU is licensed

$ nvidia-smi -q |grep License
vGPU Software Licensed Product
    License Status                    : Licensed (Expiry: 2021-10-22 20:15:22 GMT)