Using cloud vGPUs
This is not a complete article: This is a draft, a work in progress that is intended to be published into an article, which may or may not be ready for inclusion in the main wiki. It should not necessarily be considered factual or authoritative.
This guide describes how to allocate vGPU resources to a virtual machine (VM), installing the necessary drivers and checking whether the vGPU can be used. Repository access as well as access to the vGPUs, is currently only available within Arbutus Cloud. To minimize the efforts in using vGPUs, users are recommending to rely on our CVMFS software stack as much as possible, for example for accessing and using CUDA.
Supported flavors
To use a vGPU within a VM, the instance needs to be deployed on one of the flavors listed below. The vGPU will be available to the operating system via the PCI bus. While finalizing the setup for more vGPU profiles, the only flavor accessible right now is:
- vgpu1-c18-56gb
Preparation of a VM running CentOS7
Once the VM is available, make sure to update the OS to the latest available software, including the kernel and reboot the VM to have the latest kernel running.
[root@centos7]# yum -y update && reboot
Since the proprietary nvidia drivers need to be compiled against the running kernel, the package dkms is required from the EPEL Repository
[root@centos7]# yum -y install https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm
Install the Arbutus Cloud repository, it also installs the public key the package are signed with to ensure their authenticity, since these drivers and userspace tools are carefully tested first against the infrastructure, before they are made available.
[root@centos7]# yum -y install http://repo.arbutus.cloud.computecanada.ca/pulp/repos/centos/arbutus-cloud-vgpu-repo.el7.noarch.rpm
The last step is to install the nvidia vGPU packages. The kernel module package 'nvidia-vgpu-kmod', will take a few minutes as it compiles the required kernel modules in the background.
[root@centos7]# yum -y install nvidia-vgpu-kmod nvidia-vgpu-gridd nvidia-vgpu-tools
After the successful installation, the vGPU is a now accessible and licensed.
[root@centos7]# nvidia-smi
Mon Jun 1 16:03:27 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.56 Driver Version: 440.56 CUDA Version: 10.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GRID V100D-8C On | 00000000:00:05.0 Off | 0 |
| N/A N/A P0 N/A / N/A | 560MiB / 8192MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
To check for the license status as well as other information for the vGPU.
[root@centos7]# nvidia-smi -q |less
==============NVSMI LOG==============
Timestamp : Mon Jun 1 16:06:59 2020
Driver Version : 440.56
CUDA Version : 10.2
Attached GPUs : 1
GPU 00000000:00:05.0
Product Name : GRID V100D-8C
Product Brand : Grid
Display Mode : Enabled
Display Active : Disabled
Persistence Mode : Enabled
Accounting Mode : Disabled
Accounting Mode Buffer Size : 4000
Driver Model
Current : N/A
Pending : N/A
Serial Number : N/A
GPU UUID : GPU-315b585a-a41e-11ea-a63b-4ed0221b4f99
Minor Number : 0
VBIOS Version : 00.00.00.00.00
MultiGPU Board : No
Board ID : 0x5
GPU Part Number : N/A
Inforom Version
Image Version : N/A
OEM Object : N/A
ECC Object : N/A
Power Management Object : N/A
GPU Operation Mode
Current : N/A
Pending : N/A
GPU Virtualization Mode
Virtualization Mode : VGPU
Host VGPU Mode : N/A
GRID Licensed Product
Product Name : NVIDIA vComputeServer
License Status : Licensed
IBMNPU
Relaxed Ordering Mode : N/A
PCI
Bus : 0x00
Device : 0x05
Domain : 0x0000
Device Id : 0x1DB610DE
Bus Id : 00000000:00:05.0
Sub System Id : 0x139610DE
Preparation of a VM running CentOS8
Once the VM is available, make sure to update the OS to the latest available software, including the kernel and reboot the VM to have the latest kernel running.
[root@centos8]# dnf -y update && reboot
Since the proprietary nvidia drivers need to be compiled against the running kernel, the package dkms is required from the EPEL Repository
[root@centos8]# dnf -y install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm
Install the Arbutus Cloud repository, it also installs the public key the package are signed with to ensure their authenticity, since these drivers and userspace tools are carefully tested first against the infrastructure, before they are made available.
[root@centos8]# dnf -y install http://repo.arbutus.cloud.computecanada.ca/pulp/repos/centos/arbutus-cloud-vgpu-repo.el8.noarch.rpm
The last step is to install the nvidia vGPU packages. The kernel module package 'nvidia-vgpu-kmod', will take a few minutes as it compiles the required kernel modules in the background.
dnf -y install nvidia-vgpu-kmod nvidia-vgpu-gridd nvidia-vgpu-tools
After the successful installation, the vGPU is a now accessible and licensed. To check on the status, the same nvidia-smi commands can be used as seen above for Centos7.
Preparation of a VM running Debian10
Ensure that the latest packagesare installed and the system has been booted the latest stable kernel, as dkms will request the latest one available from the debian repositories.
root@debian10:~# apt-get update && apt-get -y dist-upgrade && reboot
After a successful reboot, the system should have the latest avaible kernel running and the repository can be installed, by installing the repo package. This package does also contain the gpg key all packages are signed with.
root@debian10:~# wget http://repo.arbutus.cloud.computecanada.ca/pulp/deb/debian/pool/main/arbutus-cloud-repo_0.1_all.deb root@debian10:~# dpkg -i arbutus-cloud-repo_0.1_all.deb
The installation of the package will display a warning, since the key is directly imported (for convenience) via the packages post installation procedure.
Setting up arbutus-cloud-repo (0.1) ... Warning: apt-key should not be used in scripts (called from postinst maintainerscript of the package arbutus-cloud-repo) OK
Update the local apt cache and installation of the vGPU packages.
root@debian10:~# apt-get -y install nvidia-vgpu-kmod nvidia-vgpu-tools nvidia-vgpu-gridd
After the successful installation, the vGPU is a now accessible and licensed. To check on the status, the same nvidia-smi commands can be used as seen above for Centos7.
Preparation of a VM running Ubuntu20
in progress ...