In traditional virtualization, the hypervisor acts as a "middleman," translating requests from a Virtual Machine (VM) to the physical hardware. While this makes VMs easy to manage, it adds a layer of overhead.

Hardware Passthrough (often referred to as PCI Passthrough or Directed I/O) removes that middleman. It allows a VM to take exclusive, direct ownership of a physical hardware component—like a GPU, a Network Card (NIC), or a Fibre Channel HBA.

1. How It Works: Bypassing the Hypervisor

In a standard setup, the hypervisor "emulates" hardware. When the VM wants to send a packet, it talks to a virtual driver, which talks to the hypervisor, which finally talks to the hardware.

With Hardware Passthrough, the hypervisor "hides" the device from itself and maps it directly into the VM's address space. The guest OS uses its own native manufacturers' drivers to talk to the physical pins of the device.

2. The Engine: Intel VT-d and AMD-Vi

You can't have passthrough without hardware support from the CPU and motherboard. This is handled by the IOMMU (Input-Output Memory Management Unit).

• Intel calls it VT-d (Virtualization Technology for Directed I/O).

• AMD calls it AMD-Vi.

The IOMMU ensures that the VM can only access the memory assigned to its specific hardware device, preventing the VM from "escaping" its boundaries and crashing the host or other VMs.

3. Why Use Passthrough?

There are three primary reasons why engineers choose passthrough over standard virtualization:

A. Near-Native Performance

Because there is no "translation" happening in the software, the performance is identical to running on a bare-metal server. This is critical for High-Frequency Trading (HFT) or High-Performance Computing (HPC).

B. Access to Specialized Hardware

Some hardware is too complex to emulate. If you need a VM to perform AI/Machine Learning using an NVIDIA GPU, or if you are doing high-end video encoding, you must use passthrough so the VM can access the GPU’s specialized cores (CUDA/Tensor cores).

C. Low Latency

For networking, skipping the hypervisor’s virtual switch reduces "jitter" and latency. This is why many Exadata configurations and high-end storage arrays use passthrough for their InfiniBand or RoCE adapters.

4. The Trade-offs: What do you lose?

Hardware passthrough isn't "free." By giving a VM direct control of hardware, you lose many of the "magic" features of virtualization:

• No Snapshots: You generally cannot take a live snapshot of a VM using passthrough because the hypervisor can't "capture" the internal state of the physical hardware.

• No Live Migration: You cannot easily move the VM to another server (like vMotion) because the physical device is physically plugged into the first server.

• Device Exclusivity: Once a VM "owns" a device, no other VM (and not even the host) can use it.

5. Summary: Passthrough vs. Emulation

The Bottom Line

Hardware passthrough is the "Turbo Button" of the virtualization world. It’s the best choice when the speed of the hardware is more important than the flexibility of the VM.