Unmaintained

Warning: This repository is no longer being maintained.

I recommend that you do not use this outdated kernel, as it may contain bugs or security vulnerabilities that will not be fixed.

dxgkrnl_ubuntu

Use Ubuntu on Hyper-V VM with Microsoft GPU-P support (dxgrknl kernel).

Windows Host Requirement:

• Windows 10 21H1 or later
• Windows 11 (all builds)
• Windows Server 2022 Insider Preview Build 25246 or later

Pros:

• Full Hyper-V VM with more features than WSL2 (systemd, snapd, Hyper-V External Network, ...)
• Poor man vGPU solution: one GPU is sharable among multiple Hyper-V VMs
• GPU Encoding/Decoding works: Can use Sunshine Host with hardware accelerated stream

Cons:

• Beware that any Docker CUDA image built inside this VM, will not work with bare metal machine (libcuda.so problems)
• Only tested on Ubuntu 20.04 / 22.04
• Outdated kernel 5.10, need to be updated to the WSL2 5.15 Kernel :(
• Sunshine is experimental, and GNOME can break easily
• Hyper-V is a Type 1 Hypervisor, expect heavy CPU performance loss on your Windows Host
• Should work with AMD too, but I don't have AMD cards to test
• No way to allocated a fixed amount of VRAM to each Hyper-V VM like real vGPU solution

Instructions

1. Create a Gen 2 Hyper-V virtual machine, install Ubuntu 20.04 LTS / 22.04 LTS as normal

2. Attach GPU-P adapter to your Ubuntu VM

• From Windows Host, run Powershell as Administrator

# change ubuntu to your current vm name
$vm = "ubuntu"

# this will remove any current gpu-p adapter then reattach them all
Remove-VMGpuPartitionAdapter -VMName $vm
$gpu_list = Get-VMHostPartitionableGpu
foreach ($k in $gpu_list){
    $instance_path = $k.Name
    $instance_path
    Add-VMGpuPartitionAdapter -VMName $vm -InstancePath $instance_path
}
Set-VMGpuPartitionAdapter -VMName $vm -MinPartitionVRAM 1
Set-VMGpuPartitionAdapter -VMName $vm -MaxPartitionVRAM 11
Set-VMGpuPartitionAdapter -VMName $vm -OptimalPartitionVRAM 10
Set-VMGpuPartitionAdapter -VMName $vm -MinPartitionEncode 1
Set-VMGpuPartitionAdapter -VMName $vm -MaxPartitionEncode 11
Set-VMGpuPartitionAdapter -VMName $vm -OptimalPartitionEncode 10
Set-VMGpuPartitionAdapter -VMName $vm -MinPartitionDecode 1
Set-VMGpuPartitionAdapter -VMName $vm -MaxPartitionDecode 11
Set-VMGpuPartitionAdapter -VMName $vm -OptimalPartitionDecode 10
Set-VMGpuPartitionAdapter -VMName $vm -MinPartitionCompute 1
Set-VMGpuPartitionAdapter -VMName $vm -MaxPartitionCompute 11
Set-VMGpuPartitionAdapter -VMName $vm -OptimalPartitionCompute 10
Set-VM -GuestControlledCacheTypes $true -VMName $vm
Set-VM -LowMemoryMappedIoSpace 1Gb -VMName $vm
Set-VM -HighMemoryMappedIoSpace 32GB -VMName $vm

• For Windows Server 2022 Insider Preview need to do this:

Set-ItemProperty -Path "HKLM:\SOFTWARE\Policies\Microsoft\Windows\HyperV" -Name "RequireSecureDeviceAssignment" -Type DWORD -Value 0 -Force
Set-ItemProperty -Path "HKLM:\SOFTWARE\Policies\Microsoft\Windows\HyperV" -Name "RequireSupportedDeviceAssignment" -Type DWORD -Value 0 -Force

3. Copy Windows Host GPU Driver to Ubuntu VM:

• From Ubuntu VM:

# enable SSH
sudo apt-get update && sudo apt-get install -y openssh-server 
# note down current VM ip
ip a | grep eth0  
# prepare temp_folder
mkdir -p $HOME/temp_folder/lib && mkdir -p $HOME/temp_folder/drivers 

• From Windows Host, run Powershell as Administrator

# get currentdriverfolder from this result
Get-CimInstance -ClassName Win32_VideoController -Property *

# define your vmip, vmusername and currentdriverfolder
$vmip = "192.168.1.106"
$vmusername = "abcdefg"
$currentdriverfolder = "C:\Windows\System32\DriverStore\FileRepository\nvhdcig.inf_amd64_26476eaed29e569c"

# Use scp to copy Windows Host drivers to Ubuntu VM
scp -r C:\Windows\System32\lxss\lib "${vmusername}@${vmip}:temp_folder"
scp -r $currentdriverfolder "${vmusername}@${vmip}:temp_folder/drivers"

• From Ubuntu VM:

sudo rm -rf /usr/lib/wsl && 
sudo mkdir -p /usr/lib/wsl/lib && 
sudo cp -r $HOME/temp_folder/* /usr/lib/wsl && 
sudo chmod 555 /usr/lib/wsl/lib/* && 
sudo chown -R root:root /usr/lib/wsl && 
echo "/usr/lib/wsl/lib" | sudo tee /etc/ld.so.conf.d/ld.wsl.conf && 
sudo ldconfig && 
echo "export PATH=$PATH:/usr/lib/wsl/lib" | sudo tee /etc/profile.d/wsl.sh && 
sudo chmod +x /etc/profile.d/wsl.sh

• Repeat this step 3 if you update your current Windows Host Driver

4. Install custom dxgkrnl kernel:

• From Ubuntu VM:

# Download and install dxgrknl kernel
wget https://github.com/brokeDude2901/dxgkrnl_ubuntu/releases/download/main/linux-headers-5.10.102.1-dxgrknl_5.10.102.1-dxgrknl-10.00.Custom_amd64.deb && 
wget https://github.com/brokeDude2901/dxgkrnl_ubuntu/releases/download/main/linux-image-5.10.102.1-dxgrknl_5.10.102.1-dxgrknl-10.00.Custom_amd64.deb && 
sudo dpkg -i ./linux-headers-5.10.102.1-dxgrknl_5.10.102.1-dxgrknl-10.00.Custom_amd64.deb && 
sudo dpkg -i ./linux-image-5.10.102.1-dxgrknl_5.10.102.1-dxgrknl-10.00.Custom_amd64.deb

# Make GRUB show menu for you to choose the installed dxgkrnl
sudo sed -i "s/GRUB_DEFAULT=0/GRUB_DEFAULT=saved/g" /etc/default/grub && 
sudo sed -i "s/GRUB_TIMEOUT_STYLE=hidden/GRUB_TIMEOUT_STYLE=menu/g" /etc/default/grub && 
sudo sed -i "s/GRUB_TIMEOUT=0/GRUB_TIMEOUT=30/g" /etc/default/grub && 
sudo grep -q -F "GRUB_SAVEDEFAULT=true" /etc/default/grub || echo "GRUB_SAVEDEFAULT=true" | sudo tee -a /etc/default/grub && 
sudo update-grub && cat /etc/default/grub

5. Reboot Ubuntu VM, select the new dxgkrnl kernel, enjoy !

6. (OPTIONAL) Install nvidia-docker:

sudo apt-get update && 
sudo apt-get install -y curl &&
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - &&
curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu18.04/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list &&
sudo apt-get update && sudo apt install -y nvidia-docker2

• Perform a quick Docker test see if GPU is working

sudo docker run --rm --gpus all nvcr.io/nvidia/k8s/cuda-sample:nbody nbody -gpu -benchmark && 
sudo docker run --rm -it -v /usr/lib/wsl/lib/nvidia-smi:/usr/local/bin/nvidia-smi --gpus all nvidia/cuda:11.8.0-base-ubuntu22.04 nvidia-smi

7. (OPTIONAL) Use Hyper-V Core scheduler to avoid bugs:

• By default Windows 11 use Root scheduler, which sometimes buggy, cause extremely high CPU usage
• From Windows Host, run Powershell as Administrator

bcdedit /set hypervisorschedulertype Core

• Reboot to have effect, see article here: https://learn.microsoft.com/en-us/windows-server/virtualization/hyper-v/manage/manage-hyper-v-scheduler-types

8.1 (EXPERIMENTAL) Use Moonlight / Sunshine Host for smooth remote desktop and gaming (tested on NVENC H264 and HEVC)

• Blacklist hyperv_fb bc it is too slow (see LIS/lis-next#655)

# /etc/default/grub
GRUB_CMDLINE_LINUX_DEFAULT="modprobe.blacklist=hyperv_fb"

sudo update-grub && sudo reboot

• Setup Sunshine Host on Ubuntu VM:

wget https://github.com/LizardByte/Sunshine/releases/download/v0.15.0/sunshine.deb && sudo apt-get install ./sunshine.deb
# fix libssl for ubuntu 22.04
echo "deb http://security.ubuntu.com/ubuntu focal-security main" | sudo tee /etc/apt/sources.list.d/focal-security.list
sudo apt-get update && sudo apt-get install libssl1.1
# finally run this inside Hyper-V window (bc it has an active DISPLAY session) not on SSH
sudo mkdir -p /dev/dri && sudo mkdir -p /root/.config/sunshine & sudo sunshine
# open URL to config your Sunshine Host password, later accept the right PIN code from Moonlight client

• Setup Moonlight on Windows Host from: https://moonlight-stream.org/, set the resolution to 1024x768 and Windowed Mode

GPU Encode & Decode Engine working :)

8.2 (EXPERIMENTAL) Run OpenGL Apps Fullscreen in Moonlight / Sunshine Host to prevent gnome-shell crash

xvfb-run -s "-screen 0 1024x768x24" bash -c 'sudo mkdir -p /dev/dri && sudo sunshine & glxgears'