How does one turn off the GPU or put it to idle P8?

Guides & Tutorials Tutorials
, , ,
1

To effectively manage your NVIDIA GPU’s power state and ensure it drops to the P8 idle state, you can utilize several methods, including enabling persistence mode and using specific commands.

Here’s how to do it:

Enabling NVIDIA Persistence Mode

  1. Open a Terminal: You need to run commands in a terminal window.

  2. Enable Persistence Mode: This keeps the GPU initialized and can help manage power states more effectively. Use the following command:

    sudo nvidia-smi --persistence-mode=ENABLED

    This command ensures that the GPU remains in a low-power state when not in use, which can help it drop to P8 when idle.

Checking Current Performance State

To verify the current performance state of your GPU, run:

nvidia-smi

Look for the “Performance State” in the output. It should indicate P8 when the GPU is idle.

Managing Power States Manually

If your GPU does not drop to P8 automatically, you can manually set its performance state using nvidia-settings:

  1. Use nvidia-settings: Open the NVIDIA settings interface:

    nvidia-settings

  2. Adjust PowerMizer Settings: Navigate to the “PowerMizer” section and set the mode to “Prefer Maximum Performance” or adjust as needed.

  3. Force P8 State: If necessary, you can add specific options in your configuration file (e.g., /etc/modprobe.d/nvidia.conf) to enforce power settings:

    options nvidia NVreg_RegistryDwords="PowerMizerEnable=0x1; PerfLevelSrc=0x2222; PowerMizerLevel=0x1; PowerMizerDefault=0x1; PowerMizerDefaultAC=0x1"

Additional Considerations

  • Restarting Explorer (Windows Specific): If you’re on Windows and find that your GPU is stuck at P8, try restarting Windows Explorer. This can sometimes reset the GPU state without needing a full reboot.

  • Using Scripts: For automated management, consider creating scripts that execute these commands upon startup or when certain conditions are met (e.g., when unplugging from AC power) to ensure your GPU remains in the desired state.

By following these steps, you should be able to manage your NVIDIA GPU’s power states effectively and ensure it operates efficiently at idle.

To create a script for Proxmox that manages a virtual machine (VM) and handles GPU passthrough, you might want to automate the unbinding of the GPU from the VM when it shuts down. Below is a sample Bash script that accomplishes this task.

Proxmox VM GPU Management Script

#!/bin/bash

# Variables
VMID=100  # Change this to your VM ID
GPU_PCI_ADDRESS="0000:01:00.0"  # Change this to your GPU's PCI address

# Function to unbind GPU from the VM
unbind_gpu() {
    echo "Unbinding GPU from VM..."
    echo "$GPU_PCI_ADDRESS" > /sys/bus/pci/drivers/vfio-pci/unbind
    echo "$GPU_PCI_ADDRESS" > /sys/bus/pci/drivers/nouveau/bind  # Change to 'nvidia' if using NVIDIA drivers
    echo "GPU unbound and bound to nouveau."
}

# Check the status of the VM
VM_STATUS=$(qm status $VMID)

if [ "$VM_STATUS" == "stopped" ]; then
    unbind_gpu
else
    echo "VM is still running. Current status: $VM_STATUS"
fi

Instructions

  1. Modify Variables:

    • Change VMID to the ID of your VM.
    • Update GPU_PCI_ADDRESS with the actual PCI address of your GPU.

  2. Save the Script:

    • Save the script as manage_gpu.sh on your Proxmox server.

  3. Make it Executable:

    chmod +x manage_gpu.sh

  4. Run the Script:

    • Execute the script manually or set it up as a cron job to run at specific intervals.

Automating with Hooks

To automate this process further, you can use Proxmox’s hook scripts feature:

  1. Create a Hook Script:
    Save the above script in /etc/pve/qemu-server/<VMID>.conf as a hook script that runs on shutdown.

  2. Add Hook Configuration:
    Edit the VM configuration file and add:

    hookscript: local:snippets/manage_gpu.sh

This setup will ensure that every time the specified VM is shut down, the GPU will be unbound and put into a lower power state automatically.

Note

  • Always test scripts in a safe environment before deploying them in production.
  • Ensure you have appropriate permissions and backups before modifying system files or configurations.