A global hierarchical Z space algorithm for cluster parallel graphics architectures

In this paper we present a new global hierarchical Z-space sort-last algorithm for cluster parallel graphics architectures that improves upon algorithms used so far for high performance super-graphics. The new algorithm bypasses limitations of sort-last tile based parallelization paradigms, and solves some known Z-space parallelization inefficiencies. The algorithm is implemented as a global hierarchical-Z system which allows GPUs to perform high frequency global intra-frame Z-culling and distributed final frame Z-determination. The new algorithm allows for full one-to-one process-GPU coupling with minimal inter-process and inter-GPU communications. This enables maximal input bandwidth, maximum GPU utilization levels, near optimal load balances and improved efficiency when scaled to larger configurations.