Instant Global Illumination on the GPU using OptiX Ricardo Marques and Lu´ ıs Paulo Santos Universidade do Minho, Braga, Portugal, ricjmarques@gmail.com, psantos@di.uminho.pt Abstract. OptiX, a programmable ray tracing engine, has been recently made available by NVidia, relieving rendering researchers from the id- iosyncrasies of efficient ray tracing programming and allowing them to concentrate on higher level algorithms, such as interactive global illu- mination. This paper evaluates the performance of the Instant Global Illumination algorithm on OptiX as well as the impact of three differ- ent optimization techniques: imperfect visibility, downsampling and in- terleaved sampling. Results show that interactive frame rates are indeed achievable, although the combination of all optimization techniques leads to the appearance of artifacts that compromise image quality. Sugges- tions are presented on possible ways to overcome these limitations. Keywords: instant global illumination, ray tracing, graphics processors 1 Introduction Interactive ray tracing became possible along the last decade on both CPU and GPU based platforms. However, this has been achieved through extensive optimization of code and data structures, thus developing such a ray tracer is a complex and time consuming task. In September 2009 Nvidia launched a programmable ray tracing engine for their GPUs, OptiX [1], which allows researchers to concentrate on higher level algorithms while still being able to trace rays efficiently. The goal of this paper is to assess the performance of an interactive global illumination (GI) algorithm on OptiX. This algorithm, referred to as Instant Global Illumination [2], computes indirect diffuse interreflections by generating a particle based approximation of this illumination component, resulting in a three-dimensional distribution of secondary virtual point light (VPL) sources. The algorithm in its original form is barely interactive due to the high number of VPLs. Optimizations have been proposed under the form of imperfect visibility [3], downsampling the indirect diffuse evaluation rate [4] and interleaving VPL sampling patterns [5]. This paper assesses the performance achieved with these three optimizations on a last generation NV 480 GTX GPU using OptiX and proposes a few hypothesis for further performance gains. The next section presents related work and details on the optimization tech- niques. Section 3 briefly introduces OptiX, while the used algorithm is detailed