GPU Virtualization: An overview

Virtualization is widely used in server/cloud environments to optimize resource usage and to guarantee availability. Nowadays, almost all CPUs provide hardware extensions that allow efficient virtualization of the CPU. However, with increasing use of GPUs for all kinds of parallel computations, being able to share the GPU between virtual machines is also essential for good performance.

Unfortunately, complex software stacks and lack of standardization make GPU virtualization quite challenging. Different approaches exist that insert the virtualization layer either on top of standard APIs or directly on top of the hardware. In this presentation, I want to provide an overview of these approaches and will explain why all these approaches are still relevant today.

(Hauptseminarvortrag)