A method and system are described herein for an optimization technique on two aspects of thread scheduling and dispatch when the driver is allowed to pick the scheduling attributes. The present techniques rely on an enhanced GPGPU Walker hardware command and one dimensional local identification generation to maximize thread residency.