Computer Science Department
School of Computer Science, Carnegie Mellon University
A New Architecture for Cloud Rendering
High quality graphics and realism are essential features of modern games. Until recently, players needed to own expensive consoles or outfit their PCs with the latest hardware to play games with cutting-edge graphics. This has changed with the introduction of "cloud gaming" companies, which require only a good internet connection and minimal hardware to play the latest games. These ventures suggest that the gaming industry is moving to a cloud-based model, in which computation-intensive rendering and simulation is offloaded to remote servers and only the resulting images are streamed back to the client. However, present cloud gaming services still implement a one-console-peruser model, in which across-user computation, or computation that is independent of any particular user, is repeated for all users, rather than computed only once and then shared among all users in the same scene. We present a new architecture which amortizes the cost of across-user rendering, simulation, and memory. This architecture frees additional resources for rendering and allows for improved graphics at no additional hardware cost. To analyze the performance and generality of this architecture we implemented two applications: a light field and a fluid simulation. We also devised a general means of measuring the performance of amortized algorithms in this architecture.