Announcing GPU acceleration for Indigo Renderer

Announcing GPU acceleration for Indigo Renderer

Written Friday 21 May

Graphics processing units (GPUs) have come a long way since their inception, when they were used to accelerate basic rasterisation features in high-end CAD workstations. Through standards such as 3Dfx's Glide, and later Direct3D and OpenGL, the market for inexpensive consumer GPUs has bloomed into the success story it is today.

Together with phenomenal performance increases, we have benefitted from increased programmability through new standards such as CUDA and OpenCL. It is these improvements that have allowed us to reach the stage where GPUs can be used to accelerate general purpose computations. Rendering naturally benefits from both the specialised graphics functionality and the general purpose computational power found in GPUs, and so one could say it is only a matter of time before these additional resources are exploited by all rendering software.

Today we are proud to announce GPU acceleration support for Indigo Renderer.

After considerable development effort in which we evaluated both CUDA and OpenCL platforms, we are ready to present preliminary results and ask users to join our GPU acceleration beta testing group. Both ATI and NVIDIA users are welcome to participate!

If you are interested in joining the beta group, please email us at

About the technology

We are employing a hybrid CPU+GPU approach, whereby we offload some of the CPU's computations onto the GPU. This has a number of benefits, the foremost of which being that we support the long list of features that make Indigo the powerful rendering engine it is.

Motion blur:

teapot GPU
GPU-accelerated unbiased motion blur. Images rendered in 3 minutes. Click for high-res image.


instancing test
50,000 instances and 54 billion triangles. Brute force SSS with Buddhas! Click for high-res image.

It's fast:

Image rendered in 2 minutes. Model by Ronen Bekerman. Click for high-res image.

We have also improved our path tracing code to be more intelligent and efficient, meaning that the time to produce a clean image is greatly reduced compared to other path tracers, including Indigo 2.2.

With the core functionality in place and promising first results, we will be spending time to optimise our CPU-based code to deliver awe-inspiring performance in the CPU+GPU combo; we expect more large performance increases in the near term.

All trademarks are copyright their respective owners.