This document presents a memory-efficient GPU-based ray casting algorithm for unstructured grids, significantly improving performance by using an off-the-shelf NVIDIA graphics card. The new algorithm implements a restructuring of data structures and exploits ray coherence, allowing it to use 50% less memory than previous methods while handling larger datasets. The implementation offers notable speed enhancements, achieving interactive frame rates for rendering 3D volumetric data compared to earlier CPU-based approaches.