Management of graphics processing units in a cloud platform

A computing system that controls the rendering of images by graphics processing unit (“GPU”) servers having graphics processing units, comprising:

• one or more computer-readable mediums storing computer-executable instructions that:
• receive a specification of a render task to be performed to render an image;
• identify a virtual machine of a GPU server, the virtual machine having access to a virtual graphics processing unit that is shared among multiple virtual machines hosted by the GPU server;
• direct the identified virtual machine to render the image as specified by the render task;
• and
• upon detecting an error in rendering the image, identify another virtual machine executing on another GPU server that has access to a virtual graphics processing unit that is shared among multiple virtual machines and direct the other virtual machine to render the image as specified by the render task;
• and
• a processor for executing the computer-executable instructions stored in the one or more computer-readable mediums.

A method for rendering of images in a cloud computing platform having graphics processing unit (“GPU”) servers with commodity graphics processing units, the method comprising:

• allocating to a program a pool of virtual machines for executing on the GPU servers, the virtual machines interfacing with a virtual graphics processing unit;
• receiving from the program a specification of a render task to be performed to render an image;
• identifying virtual machines from the pool of virtual machines to perform the render task, the identified virtual machines executing on different GPU servers with commodity graphics processing units;
• for each of the identified virtual machines, directing that virtual machine to render the image as specified by the render task such that each identified virtual machine is directed to render the same image;
• and
• upon first detecting that a virtual machine has successfully rendered the image, provide to the program the image rendered by that virtual machine,
• such that multiple virtual machines are directed to render the same image to increase the chances of the image being successfully rendered even though some of the virtual machines may not successfully render the image.