Mitsuba is a research-oriented rendering system in the style of PBRT, from which it derives much inspiration. It is written in portable C++, implements unbiased as well as biased techniques, and contains heavy optimizations targeted towards current CPU architectures. Mitsuba is extremely modular: it consists of a small set of core libraries and over 100 different plugins that implement functionality ranging from materials and light sources to complete rendering algorithms.

In comparison to other open source renderers, Mitsuba places a strong emphasis on experimental rendering techniques, such as path-based formulations of Metropolis Light Transport and volumetric modeling approaches. Thus, it may be of genuine interest to those who would like to experiment with such techniques that haven't yet found their way into mainstream renderers, and it also provides a solid foundation for research in this domain.

The renderer currently runs on Linux, MacOS X and Microsoft Windows and makes use of SSE2 optimizations on x86 and x86_64 platforms. So far, its main use has been as a testbed for algorithm development in computer graphics, but there are many other interesting applications.

Mitsuba comes with a command-line interface as well as a graphical frontend to interactively explore scenes. While navigating, a rough preview is shown that becomes increasingly accurate as soon as all movements are stopped. Once a viewpoint has been chosen, a wide range of rendering techniques can be used to generate images, and their parameters can be tuned from within the program.

Mitsuba instances can be merged into large clusters, which transparently distribute and jointly execute tasks assigned to them using only node-to-node communcation. It has successfully scaled to large-scale renderings that involved more than 1000 cores working on a single image. Most algorithms in Mitsuba are written using a generic parallelization layer, which can tap into this cluster-wide parallelism. The principle is that if any component of the renderer produces work that takes longer than a second or so, it at least ought to use all of the processing power it can get.

The renderer also tries to be very conservative in its use of memory, which allows it to handle large scenes (>30 million triangles) and multi-gigabyte heterogeneous volumes on consumer hardware.