The proposed LOD system operates within the ray-tracing acceleration structure, employing a residency map that aligns with the bounding volume hierarchy of scene objects. This enables the graphics processing unit (GPU) to ascertain the required quality level for objects at any given moment, thus optimizing memory usage.
In traditional ray-tracing implementations, which lack a dynamic LOD system, the quality of ray-traced effects is constant, leading to high memory consumption and, consequently, a performance bottleneck. To mitigate this, developers often resort to upscaling technologies like DLSS, FSR, XeSS, or console-specific upscaling methods, which effectively camouflage the performance impact of ray tracing. However, an LOD system dedicated to the ray-tracing pipeline could offer developers greater flexibility in managing performance, irrespective of the rendering resolution.
The practical implications of this technology are significant, potentially extending the viability of GPUs with limited VRAM in rendering modern games with ray tracing enabled. This could be particularly beneficial for GPUs with 8GB or 10GB of VRAM, which may otherwise struggle with frame pacing in ray-traced games due to memory constraints. Furthermore, this LOD system might enable smoother ray-tracing performance on GPUs with as little as 4GB to 6GB of VRAM, such as the mobile RTX 3050 series, and could also enhance ray-tracing capabilities on gaming consoles like the PlayStation 5, where available memory for games is around 12GB out of the total 16GB.
Source: tomshardware