GPU passthrough

noun

Graphics Processing Unit passthrough

GPU passthrough is a virtualization technique that allows a virtual machine (VM) to access a physical graphics processing unit (GPU) directly, bypassing the host system’s operating system. This is done by assigning the entire GPU to the VM, making it behave as if it were connected to a real machine.

It works by using hardware virtualization features like Intel VT-d or AMD-Vi to isolate the GPU from the host system. The VM can then use the GPU’s full capabilities, which is crucial for tasks like 3D rendering, gaming, or machine learning. However, it requires compatible hardware and software, and the GPU cannot be used by the host system while assigned to the VM.

For users, GPU passthrough changes the way they can run demanding applications in virtualized environments. It enables near-native performance for graphics-intensive tasks, but it also means the host system loses access to the GPU. For example, a gamer running a VM with GPU passthrough can play modern games at full settings, but the host system will have no GPU resources left for other tasks.

A data scientist uses GPU passthrough to run a deep learning model in a VM, achieving the same speed as on a bare-metal machine.

See also: Hardware acceleration

Découvrez nos autres services

ElpisIA, c’est tout un univers — explorez nos autres services.