Abstract: With the advancements in realtime ray tracing and new global illumination algorithms we are now able to render the most important illumination effects at interactive rates. One of the major remaining issues is the fast and ef cient simulation of caustic illumination, such as e.g. the illumination from a car headlight. The photon mapping algorithm is a simple and robust approach that generates high-quality results and is the preferred algorithm for computing caustic illumination. However, photon mapping has a number of properties that make it rather slow on today's processors. Photon mapping has also been notoriously difficult to parallelize efficiently. In this paper, we present a detailed analysis of the performance issues of photon mapping together with significant performance improvements for all aspects of the photon mapping technique. The solution forms a complete framework for realtime photon mapping that efficiently combines realtime ray tracing, optimized and improved photon mapping algorithms, and efficient parallelization across commodity PCs. The presented system achieves realtime photon mapping performance of up to 22 frames per second on non-trivial scenes, while still allowing for interactively updating all aspects of the scene, including lighting, material properties, and geometry.