Dynamic and application-specific virtualized graphics processing

A computer-implemented method, comprising:

• initiating execution of an application on a virtual compute instance attached to a virtual graphics processing unit (GPU) accessible to the virtual compute instance over a network using encapsulated graphics API commands of a remoting protocol, wherein the virtual GPU is implemented using a physical GPU resource allocated from a pool of available physical GPU resources;
• detecting an idle state of the virtual GPU, and responsive to detecting the idle state:
• detaching the virtual GPU from the virtual compute instance;
• and
• returning the physical GPU resource allocated from the pool of available physical GPU resources to the pool of available physical GPU resources.

A computer-readable storage medium storing program instructions computer-executable to perform:

• initiating execution of an application on a virtual compute instance attached to a virtual graphics processing unit (GPU) accessible to the virtual compute instance over a network using encapsulated graphics API commands of a remoting protocol, wherein the virtual GPU is implemented using the physical GPU resource allocated from a pool of available physical GPU resources;
• detecting an idle state of the virtual GPU, and responsive to detecting the idle state:
• detaching the virtual GPU from the virtual compute instance the virtual GPU;
• and
• returning the physical GPU resource allocated from the pool of available physical GPU resources to the pool of available physical GPU resources.