Microsoft Proposes LOD-Style Ray Tracing Calculations To Save On VRAM

Microsoft Proposes LOD-Style Ray Tracing Calculations To Save On VRAM

While ray tracing provides incredibly detailed fidelity to games and 3D applications, one drawback is the resource cost. It’s extremely demanding on compute and memory, which is why all GPU vendors today provide some sort of upscaling solutions to compensate for performance loss. However, Microsoft’s patent is looking to solve at least part of that problem.

Games with ray tracing demand lots of VRAM on tap, as the BVH (bounding volume hierarchy) calculations require huge amounts of memory to work. The patent, as Microsoft puts it, relies on a model similar to how LOD (level of detail) works in a typical 3D environment.

So what exactly is LOD? In a 3D environment, developers can save on compute demands by lowering the quality of the object when it’s far enough from the player’s view that the difference in render quality is imperceptible. In such cases, only objects very near the field of view are needed to render at their highest detail, which can save resources for extra performance. Similarly, in Microsoft’s patent – the further the light source is, the less compute and memory is required to produce similar levels of visual quality.

Today’s ray tracing implementation is largely ‘brute-force’ style calculations, especially with path tracing implementations where large amounts of rays are required to render a realistic scene. While NVIDIA has overcome this with denoising, ray reconstruction, and upscaling at the end of the render pipeline, Microsoft’s solution likely has some pretty significant performance potential as well.

Source: Tom’s Hardware

Pokdepinion: Hopefully this means games with RT will be less demanding moving forward.