Standard OpenGL-based rendering has sampling limitations. By default these rendering systems point sample rendered pixels. For highly resolved objects, this sampling is adequate to represent the object accurately, but when the object has a relatively small projected area that is on the order of a few pixels, the object intensity is corrupted with aliasing. Hardware anti-aliasing such as multisampling provides minimal relief by offering 4, 8, or 16 samples within a single pixel. However, for hardware-in-the-loop (HITL) scene generation where accurate energy conservation of unresolved sub-pixel objects must be maintained, standard hardware anti-aliasing is not good enough. Zoom anti-aliasing (ZAA) has been proven as a viable solution for rendering objects that would otherwise be grossly under-sampled. Techniques in the past have focused on processing the zoom window pixels in the CPU because the graphics processor unit (GPU) was not general purpose enough to support the zoom window processing. However, this is no longer the case because of the new capabilities of modern graphics processors. This paper presents a modern GPU-based zoom window approach and compares the results to a classic CPU-based approach.