Spatial Partitioning Strategies for Memory-Efficient Ray Tracing of Particles

Untitled

“…Nearly a decade later, with the surge of Artificial Intelligence (AI), the community realized that the performance of GPUs was not high enough to properly handle the new Deep Learning models being developed. For this reason, near 2017, NVIDIA introduced tensor cores [3][4][5][6][7][8][9][10][11][12] inside the chip to further accelerate the performance of all AI applications. GPU Tensor cores are Application Specific Integrated Circuits (ASICs), or simply specific-purpose cores that perform fast matrix multiply accumulate (MMA) operations.…”

“…Doing it by brute force would mean checking all triangles of the scene for each ray, making it very inefficient. Space partitioning trees [9] and other variants of trees have been implemented in GPU [8], although the nature of trees introduce a difficult irregular memory accesses for the GPU architecture which is limited in this aspect. As a solution to this problem, an RT core offers a hardware implemented Bounding Volumne Hierarchy (BVH) tree data structure [15], allowing a ray to find ray/triangle intersections (other custom primitives as well) overall significantly faster than the software-implemented alternatives.…”

“…Successful research has been done in the recent years. In the case of tensor cores, new ways have been proposed to further accelerate arithmetic reductions [16,13,[5][6][7][8][9][10][11][12][17][18][19][20][21] prefix sum [4-12, 17-21, 22-29] Fast Fourier Transform [22], [10], [23], [5], stencil computations for PDE simulations [11] and even fractals [14,. In general, all of these works achieve significant higher performance when compared to doing it traditionally in GPU.…”

Probabilistic Occlusion Culling using Confidence Maps for High-Quality Rendering of Large Particle Data

“…This approach also requires an acceleration structure to be efficient, which can only have two levels, of which the bottom one has bounding information [35]. A more recent approach investigates nesting of P-k-ds inside a standard BVH to balance memory requirements and resulting performance [12] and benchmarks implementations using OptiX, OSPRay and OpenGL.…”

“…Ray tracing. We compare our method against GPU ray tracing using P-k-d trees [12] in App. D. While P-k-d trees are faster per sample, increasing the number of samples per pixel to allow for sub-pixel detail degrades performance linearly, as expected.…”