Multithreaded rendering for virtualized graphics processing

A computer-implemented method, comprising:

• receiving a plurality of commands from a computing device at a graphics processing unit (GPU) server via a network, wherein a virtual GPU is implemented using a physical GPU of the GPU server and attached to the computing device, wherein the commands are generated by a plurality of application threads executing in parallel and associated with a single virtual compute instance on the computing device, and wherein individual ones of the commands are associated with identifiers of individual ones of the application threads;
• assigning individual ones of the commands to individual ones of a plurality of execution threads on the GPU server, wherein the individual ones of the execution threads correspond to the individual ones of the application threads, and wherein the commands are assigned based at least in part on the identifiers;
• and
• generating GPU output using the commands based at least in part on parallel execution of at least some of the execution threads.

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

• receiving a plurality of commands from a compute instance at a graphics processing unit (GPU) server via a network, wherein a virtual GPU is implemented using a physical GPU of the GPU server and attached to the compute instance, wherein the commands are generated by a plurality of application threads executing in parallel and associated with a single virtual compute instance on the compute instance, and wherein individual ones of the commands are associated with identifiers of individual ones of the application threads;
• assigning individual ones of the commands to individual ones of a plurality of execution threads on the GPU server, wherein the individual ones of the execution threads correspond to the individual ones of the application threads, and wherein the commands are assigned based at least in part on the identifiers;
• and
• generating GPU output based at least in part on parallel execution of the commands using the execution threads.