Delivering GPU resources across machine boundaries

摘要

Described herein is providing GPU resources across machine boundaries. Data centers tend to have racks of servers that have limited access to GPUs. Accordingly, disclosed herein is providing GPU resources to computing devices that have limited access to GPUs across machine boundaries.

主权项

1. A method for providing graphics processing unit (GPU) resources, the method comprising:
processing a request from an application operating on a virtual machine comprising a first system to perform at least one GPU operation; sending data associated with the performance of at least one GPU operation to a Virtual graphics driver on the virtual machine to a separate system comprising a GPU for processing of the data by the GPU on the separate system, wherein the separate system mounts a state corresponding to the virtual machine and maintains the state between two or more GPU processing tasks, the GPU processing tasks comprising at least one operation; receiving processed data from the separate system after the at least one GPU operation is performed; and providing an indication to the application that the at least one GPU operation is completed.