VGPU Unlock

2022-03-14T20:47:28Z

Krutav: Add page

vGPU_Unlock is a project created by Jonathan Johansson, available on [https://github.com/DualCoder/vgpu_unlock GitHub].

The project is designed to allow Nvidia's enterprise GPU virtualization technology to run on consumer Nvidia graphics cards (ex: GeForce GTX series).

Since its inception, there have been several forks of the project that are more up to date and more performant (ex: Rust version).

Documentation for the original project is available [https://docs.google.com/document/d/1pzrWJ9h-zANCtyqRgS7Vzla0Y8Ea2-5z2HEi4X75d2Q/edit?usp=sharing here].

GPU Support

2022-03-14T20:37:12Z

Krutav: Added some more information for vGPU

== Abstract ==
The following page document will detail GPU Support for SR-IOV and VFIO-Mdev functionality.

== Overview ==
{| class="wikitable"
|+Feature Matrix
!Vendor
!Architecture
!Model
!Libre Drivers
!Partially Libre Drivers (binary blobs)
!Non-Libre Drivers
!Merged Host+Guest Acceleration
!Vendor Disables Hardware Memory Protection (SR-IOV)
!VFIO-mdev support
!LibVF.IO Support
!OpenMdev Support (arcd-ddmam)
|-
|Intel
|6th Generation (Skylake)
|*HD
|i915
|
|
|Yes
|Yes
|Yes
|Yes
|Yes
|-
|Intel
|7th Generation (Kaby Lake)
|*HD
|i915
|
|
|Yes
|Yes
|Yes
|Yes
|Yes
|-
|Intel
|8th Generation (Coffee Lake)
|*HD
|i915
|
|
|Yes
|Yes
|Yes
|Yes
|Yes
|-
|Intel
|9th Generation (Cannon Lake)
|*HD
|i915
|
|
|Yes
|Yes
|Yes
|Yes
|Yes
|-
|Intel
|10th Generation (Ice Lake)
|
|i915
|
|
|No
|Yes
|No
|No
|No
|-
|Intel
|11th Generation (Tiger Lake)
|Xe*
|i915
|
|
|
|No
|
|Yes
|
|-
|Intel
|12th Generation
(Alder Lake)
|Xe*
|i915
|
|
|
|No
|
|Yes
|
|-
|Nvidia
|Maxell
|*(see exceptions below)
|Nouveau
|
|Consumer Proprietary, Proprietary Mdev
|Yes
|Yes
|Yes
|Yes
|Yes
|-
|Nvidia
|Pascal
|*
|Nouveau
|
|Consumer Proprietary, Proprietary Mdev
|Yes
|Yes
|Yes
|Yes
|Yes
|-
|Nvidia
|Turing
|*
|Nouveau
|
|Consumer Proprietary, Proprietary Mdev
|Yes
|Yes
|Yes
|Yes
|Yes
|-
|Nvidia
|Ampere
|A*
|Nouveau (In Development)
|
|Consumer Proprietary, Proprietary Mdev
|
|No
|Yes
|Yes
|Yes (Unreleased)
|-
|Nvidia
|Ampere
|RTX 30**
|Nouveau (In Development)
|
|Consumer Proprietary, Proprietary Mdev
|
|Yes
|Yes
|Yes
|Yes (Unreleased)
|-
|AMD
|Tonga
|W7100
|AMDGPU
|GPU-IOV Module (Deprecated)
|AMDGPU-Pro
|No
|Yes ([https://forum.level1techs.com/t/how-to-sr-iov-mod-the-w7100-gpu/164186 patched])
|No
|Yes
|Yes (Prototype)
|-
|AMD
|Tonga
|S71**
|AMDGPU
|GPU-IOV Module (Deprecated)
|AMDGPU-Pro
|No
|No
|No
|Yes
|Yes (Prototype)
|-
|AMD
|Vega
|v340
|AMDGPU
|
|AMDGPU-Pro, GPU-IOV Module
|No
|No
|Unconfirmed
|Yes
|No
|-
|AMD
|Navi
|v520
|AMDGPU
|
|AMDGPU-Pro, GPU-IOV Module
|No
|No
|Unconfirmed
|Yes
|No
|}

== Intel ==
This section will cover Mdev support on Intel GPUs.

==== Hardware ====
GPUs supporting GVT-g (software based mediation)[https://www.intel.com/content/www/us/en/support/articles/000058558/graphics.html]:

* Intel HD Graphics 5500
* Intel HD Graphics 6000
* Intel HD Graphics 510
* Intel HD Graphics 520
* Intel HD Graphics 530
* Intel HD Graphics 620
* Intel HD Graphics 630
* Intel UHD Graphics 620
* Intel UHD Graphics 630

GPUs supporting SR-IOV (hardware based mediation):

* Intel Iris Xe Graphics
* Intel Iris Xe MAX Graphics

==== Drivers ====
Drivers with GVT-g support:

* i915

== Nvidia ==
This section will cover Mdev support on Nvidia GPUs.

==== Hardware ====
GPUs with official support for vGPU (Source: [https://docs.nvidia.com/grid/gpus-supported-by-vgpu.html]):

* NVIDIA A100 HGX 80GB
* NVIDIA A100 PCIe 80GB (SR-IOV: 20VF [https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/PB-10577-001_v02.pdf])
* NVIDIA A100X
* NVIDIA A100 HGX 40GB
* NVIDIA A100 PCIe 40GB (SR-IOV: 16VF [https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/A100-PCIE-Prduct-Brief.pdf])
* NVIDIA A40 (SR-IOV: 32VF [https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a40/NVIDIA%20A40%20Product%20Brief.pdf])
* NVIDIA A30 (SR-IOV: 8VF [https://www.nvidia.com/content/dam/en-zz/Solutions/data-center/products/a30-gpu/pdf/a30-product-brief.pdf])
* NVIDIA A30X
* NVIDIA A16 (SR-IOV: 16VF [https://images.nvidia.com/content/Solutions/data-center/vgpu-a16-product-brief.pdf])
* NVIDIA A10 (SR-IOV: 32VF [https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a10/pdf/A10-Product-Brief.pdf])
* NVIDIA A2 (SR-IOV: 16VF [https://www.nvidia.com/content/dam/en-zz/solutions/data-center/a2/pdf/a2-product-brief.pdf])
* NVIDIA RTX A6000
* NVIDIA RTX A5000
* Quadro RTX 8000 (SR-IOV: 24 VF [https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/quadro-product-literature/NVIDIA-Quadro-RTX-8000-PCIe-Server-Card-PB-FINAL-1219.pdf])
* Quadro RTX 8000 passive
* Quadro RTX 6000 (SR-IOV: 24 VF [https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/quadro-product-literature/NVIDIA-Quadro-RTX-6000-PCIe-Server-Card-PB-FINAL-1219.pdf])
* Quadro RTX 6000 passive
* Tesla V100
* Tesla T4 (SR-IOV: 16VF [https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/tesla-t4/t4-tensor-core-product-brief.pdf])
* Tesla P100
* Tesla P40
* Tesla P6
* Tesla P4
* Tesla M60
* Tesla M10
* Tesla M6
* GRID K2 (Citrix XenServer and VMware ESXi only)
* GRID K1 (Citrix XenServer and VMware ESXi only)

Additionally supported GPUs with [[VGPU Unlock]]:

* All(?) Volta based GPUs (V-series)
*All(?) Turing based GPUs (20-series)
* All(?) Pascal based GPUs (10-series)
* Most Maxwell based GPUs (9-series) with the following notable exceptions:
** GTX 970 (mixed FB regions)
** Quadro M2000 (mixed FB regions)
** Tesla M40 [https://github.com/DualCoder/vgpu_unlock/issues/62]

==== Drivers ====
In order to use Mdev with Nvidia GPUs, a proprietary enterprise driver is required.

The Nvidia driver supports both SR-IOV and software based mediation and chooses between them based on what the installed GPU supports.

'''Licensing'''

Usage of the proprietary enterprise driver is subject to [https://www.nvidia.com/en-us/data-center/buy-grid/ license fees].

== AMD ==
This section will cover Mdev support on AMD GPUs.

==== Hardware ====
Supported GPUs:

* AMD FirePro S7150
* AMD Radeon Pro V520 [https://www.amd.com/en/graphics/workstation-virtual-graphics]
*AMD FirePro W7100 [https://forum.level1techs.com/t/how-to-sr-iov-mod-the-w7100-gpu/164186 <nowiki>[15]</nowiki>]

==== Drivers ====
The open source [https://github.com/GPUOpen-LibrariesAndSDKs/MxGPU-Virtualization GIM driver] supports AMD Tonga architecture GPUs such as the FirePro S7150/W7100. Modern kernel versions require [https://github.com/GPUOpen-LibrariesAndSDKs/MxGPU-Virtualization/pull/24 pull request 24].

Open-IOV - User contributions [en]

VGPU Unlock

GPU Support