GPU_PASSTHROUGH_v2

DragonForge

Open Source — Enterprise GPU Passthrough

Dedicate your entire GPU to any workload — gaming, AI training, 3D rendering, or development. Stealth anti-detection, 4MB OVMF with TPM 2.0, multi-vendor GPU support. Zero manual setup.

dragonforge v2.0.0

$ sudo dragonforge setup

✓ Found GPU: RTX 4060 [10de:28a0]

✓ Config auto-generated

✓ Hook installed + VFIO configured

✓ Symlink: dragonforge → installed

$ dragonforge create

✓ OVMF 4MB firmware (TPM2 built-in)

✓ TPM state cleared for fresh TPM

✓ Stealth: SMBIOS spoofed, KVM hidden

✓ VM defined: win11-dragonforge

$ dragonforge start

▸ Pre-flight: VFIO ✓ GPU ✓ Temp 42°C ✓

✓ GPU passed through — VM running

Use Cases

GPU passthrough isn't just for gaming. Any workload that needs bare-metal GPU access in an isolated environment benefits from DragonForge.

Gaming

Play Windows-only games with kernel-level anti-cheat (BattleEye, EAC, Vanguard) that will never work on Wine/Proton. Stealth VM hides from anti-cheat detection — full GPU performance, no dual boot.

FortniteValorantDestiny 2The Crew 2RobloxEA FC

AI & Machine Learning

Isolate CUDA workloads in dedicated VMs. Run training jobs, inference servers, or Jupyter notebooks with full GPU access while your host stays responsive.

PyTorch trainingTensorFlowLLM inferenceStable DiffusionComfyUICUDA dev

3D & Creative

GPU-accelerated rendering in isolated environments. Run Blender, DaVinci Resolve, or CAD tools in a Windows VM with bare-metal GPU performance.

Blender CyclesDaVinci ResolveAdobe SuiteUnreal EngineUnityAutoCAD

Dev & Testing

Spin up GPU-enabled VMs for CUDA development, driver testing, or multi-OS builds. Test on real Windows without leaving Linux.

CUDA devDriver testingCI/CD GPU jobsCross-platform buildsGPU benchmarkingVulkan dev
// v2.0.0

What's New

DragonForge v2 ships stealth anti-detection for gaming anti-cheat, 4MB OVMF firmware with native TPM 2.0 drivers, multi-vendor GPU support, and SELinux-resilient swtpm management.

4MB OVMF firmware with built-in TPM2 drivers
Stealth anti-detection — hidden hypervisor, spoofed SMBIOS
Multi-vendor GPU support (NVIDIA + AMD)
SELinux-resilient swtpm lifecycle management
Randomized VM serial numbers and board IDs
Pre-flight VFIO + GPU + thermal checks
Emergency GPU reset and watchdog timeout
Clean TPM recreation on every vm-create

How It Works

Zero-Config Setup

Auto-detects your GPU, finds ISOs, generates config, installs the libvirt hook, and configures VFIO. One command from clone to running.

Stealth Anti-Detection

Hides VM fingerprints from anti-cheat systems. Spoofed SMBIOS, randomized serials, hidden hypervisor leaf, KVM stealth flags.

Secure Boot + TPM 2.0

4MB OVMF firmware with native TPM 2.0 drivers. Windows 11 installs without workarounds. Proper tpm-crb interface with swtpm.

Hot-Toggle GPU

Switch your GPU between host and VM without rebooting. Device-level unbind keeps your desktop compositor alive.

Multi-Vendor GPU

Full support for NVIDIA and AMD GPUs. Vendor-specific hooks for driver unbind/rebind, ROM bar handling, and PCI reset.

Hardware Health

GPU temperature checks, battery-aware warnings on laptops, 30-second watchdog prevents hook deadlocks.

Near-Native Performance

KVM/QEMU with CPU topology, VirtIO, and IOMMU gives 95-98% bare-metal performance. Your GPU runs at full power.

SELinux Resilient

Clean swtpm lifecycle management. Stale socket cleanup prevents SELinux label conflicts. Works in enforcing mode.

Self-Healing

Timeout watchdog, pre-flight checks on every start, emergency GPU reset. Clean TPM recreation on every vm-create.

Requirements

CPU

Intel VT-d or AMD-Vi (IOMMU support)

GPUs

2 GPUs — integrated (iGPU) + discrete NVIDIA/AMD

IOMMU

GPU must be in its own IOMMU group

RAM

16GB+ recommended (split between host and VM)

Storage

120GB+ free for VM disk

Linux

Fedora, Arch, Ubuntu 22.04+ (KVM support)

Your GPU, Your Rules

Keep Linux as your daily driver. Dedicate your GPU to any workload — gaming with stealth anti-cheat bypass, AI training, rendering. Three commands from clone to running.