A Single (Unified) Shader GPU Microarchitecture for Embedded Systems

We present and evaluate the TILA-rin GPU microarchitecture for embedded systems using the ATTILA GPU simulation framework. We use a trace from an execution of the Unreal Tournament 2004 PC game to evaluate and compare the performance of the proposed embedded GPU against a baseline GPU architecture for the PC. We evaluate the different elements that have been removed from the baseline GPU architecture to accommodate the architecture to the restricted power, bandwidth and area budgets of embedded systems. The unified shader architecture we present processes vertices, triangles and fragments in a single processing unit saving space and reducing hardware complexity. The proposed embedded GPU architecture sustains 20 frames per second on the selected UT 2004 trace.