From a recent announcement:

Glare Technologies is proud to announce the release of Indigo Renderer 3.0 and Indigo RT. We use a hybrid GPU acceleration approach, which typically results in a 2-3x speedup when paired with a sufficiently powerful CPU. Realtime scene changes are possible, also in conjunction with network rendering to further accelerate rendering performance. A page outlining the other features and improvements of Indigo 3.0 and Indigo RT can be found at http://www.indigorenderer.com/indigo3 and http://www.indigorenderer.com/indigo_rt.

The new MachStudio Pro is stand-alone, visualization and rendering software that uses multi-threaded GPGPU computing to enable 3D artists to create and manipulate lights, materials, and HDR cameras in a real-time, non-linear workflow environment with film quality results.

This paper by Robert et al. at the University of Bern, Switzerland describes the object intersection buffer (OIB), a GPU-based visibility preprocessing algorithm for accelerating ray tracing. Based on this approach, a hybrid ray tracer is proposed to exploit parallel ray tracing using the GPU and CPU. (Hybrid Ray Tracing – Ray Tracing Using GPU-Accelerated Image-Space Methods. Philippe C.D. Robert, Severin Schoepke, and Hanspeter Bieri. Proceedings of GRAPP 2007.)

Neoptica has recently posted a whitepaper, “Programmable Graphics—The Future of Interactive Rendering.” It introduces the coming era of programmable graphics, in which developers implement rendering algorithms using combinations of parallel CPU and GPU tasks executing cooperatively on heterogeneous multi-core architectures of the near future. By embracing both task- and data-parallel computation, this approach frees developers to use the most efficient parallel computation style for their algorithms, and makes it possible to define custom graphics pipelines built using complex algorithms and dynamic data structures. The paper argues that future graphics applications that leverage the tightly coupled capabilities of forthcoming CPUs and GPUs will generate far richer and more realistic imagery, use computational resources more efficiently, and scale to large numbers of CPU and GPU cores.