OpenSubdiv 3.0 coming (Alpha update)


The alpha release has just been made public (Sept. 5th). All the new code was moved into the public github repository so that anyone can access it and download. The documentation is available here and is also updated.

Disclaimer from the team: “it’s an alpha release, so it’s not 100% complete yet, but there is enough in place to give a pretty good idea of where things are going. The feature-complete Beta is currently slated for “late Q4 2014”.

One important discussion that happened at SIGGRAPH that we have not had a chance to report about is the updates to the OpenSubdiv project. There was a range of interesting news from the conference.

OpenSubdiv is a set of open source libraries that implement high-performance subdivision surface (subdiv) evaluation on massively parallel CPU and GPU architectures. The resulting limit surface matches Pixar’s RenderMan to numerical precision. The code embodies decades of research and experience by Pixar, and a more recent and still active collaboration on fast GPU drawing between Microsoft Research, Pixar, and others such as DreamWorks Animation.

DreamWorks Animation announced at SIGGRAPH in Vancouver that their contribution to Pixar’s OpenSubdiv project would be included and become part of OpenSubdiv version 3.0 when it is released later in the year (2014).

The contribution refactors the front end of OpenSubdiv, creating a new low-level layer that separates the mathematics of subdivision from OpenSubdiv’s existing data structures and providing a simplified and accessible core for the standard. New data structures have been built upon this core to eliminate restrictions on the kinds of meshes that can be processed, significantly accelerating both general CPU processing and GPU pre-processing. The work was undertaken by DreamWorks Animation’s Principal Engineer Barry Fowler, in consultation with Pixar.

“This is a significant and important contribution to OpenSubdiv,” said Dr. Steve May, Chief Technology Officer at Pixar. “Barry Fowler’s approach not only improves the front end for easier adoption, but makes the back end faster too by directly facilitating a more efficient compute strategy for GPUs. It’s great to get such a fundamental contribution from another animation studio.”

“Pixar has taken a bold and positive step toward standardizing surface modelling with the development of OpenSubdiv,” said Dr. Lincoln Wallen, Chief Technology Officer at DreamWorks Animation. “We are excited to collaborate with Pixar by sharing our technical expertise in integrating optimizations, modifications, and re-structured code into the standard further benefiting our creative community.”

While DreamWorks was not already using OpenSubdiv for shots in films such as The Croods, they have been working with it. The model below of Grug Crood shows how such a model would appear in say Maya with OpenSubdiv 3.0 when released. If an animator is trying to do complex animation, especially say facial animation between Grug and Ugga Crood, it is vital that the maximum detail and subtle expressions be both fast and easy to control. One of the great advantages for animators when using systems with OpenSubdiv is getting a more accurate representation of the final look while animating. 

With OpenSubdiv 3.0 Bill Polson of Pixar comments, “we are accepting the biggest change since OpenSubdiv 1.0: a new set of core data structures from DreamWorks Animation Studios. Among other things, this separates subdivision math from the data representation. This allows developers to work with their current data structures, rather than having to marshal data in and out of OpenSubdiv. This will simplify many things, and should speed adoption… It’s really hard to overstate its importance. It’s faster to recalculate the topo analysis, it’s faster to draw, it’s cleaner, it handles more geo cases, and to me the best part, it allows you to have your own geo representation” he told fxguide.

Pixar has lead the charge on OpenSubdiv and the open source development. It now must be incredibly rewarding to see other studios also making serious contributions back into the program. Andrew Pearce, Director of R&D at DreamWorks Animation, is to be commended for his open source support that benefits everyone.



Autodesk also announced support in 3ds Max. This means OpenSubdiv is now standard for subdivision content creation in Maya, Modo, Blender, Houdini, Mari, Cinebox, and (soon) 3ds Max

The “new support for OpenSubdiv in 3dsMax was the top-rated feature request on User Voice,” posted Autodesk when it announced the major modeling addition. It is part of Extension 1 for 3ds Max 2015.

In particular, OpenSubdiv support in 3ds Max provides an efficiently represent subdivision surfaces using OpenSubdiv libraries, with the ability to:

  • Visualize one’s model interactively without the need to render for a better WYSIWYG experience, with more accurate realization of an artist’s creative intent.
  • Faster in-viewport performance for meshes with high subdivision levels by taking advantage of both parallel CPU and GPU architectures.
  • Reduce the guesswork when animating models and help shorten production times, with efficient smooth surface representation.
  • Create complex topology in less time, by taking advantage of efficient crease modeling workflows with CreaseSet modifier and Crease Explorer.
  • Easily transfer of models to and from certain other packages that support OpenSubdiv using Autodesk® FBX® asset exchange technology and achieve a consistent appearance.


OpenSubdiv is standard for subdivision in renderers:  RenderMan, Arnold *, Mental Ray, VRay (Note: most renderers use polygons in their inner algorithms. But these renderers can accept OpenSubdiv surfaces as input geometry).

************ UPDATE: We checked on the Arnold comment above and got some clarification from Bill Polson at Pixar after one of our readers flagged the issue (thanks): “I mentioned that Arnold had adopted, in that they have a preprocessor that will handle OpenSubdiv meshes (with semi-sharp creases etc.) and generate input data for their renderer. Turns out that status is a bit premature. They have an implementation, but after going over it at SIGGRAPH with the Pixar and DigitalFish teams, they decided to tweak it just a bit before releasing. Stay tuned to the SolidAngle guys for definitive word on when this will land.”


And there is a lot of interest (and already some adoption) outside of animation.

Work done by Digital Fish who consult extensively on OpenSubdiv pipelines

Dan Herman of DigitalFish posted : “We are finding good applications for OpenSubdiv on the digital-media side of conventional manufacturing. While CAD is still dominated by NURBS, surrounding areas — conceptual and graphic design, visualization and marketing — require creative tools for smooth, high-precision geometry at varying levels of detail. Conventional NURBS and polygon tools are often a poor fit. Some of these users are adopting OpenSubdiv…

DigitalFish and Google/Motorola have beefed up the mobile implementations. NVIDIA has improved the CUDA implementation. Intel has improved CPU multi-threading. “Further, subdivs show promise in new fields such as 3-D printing and simulation-driven design. Several major CAD vendors have recently introduced subdiv support for these applications. For these vendors, OpenSubdiv offers a well-validated, portable, performant and standards-compliant subdivision surface implementation.” he adds.

Bill Polson of Pixar adds ” OpenSubdiv is synonymous with “Catmull-Clark Subdivision Surfaces. There is no competing standard. Approximate and inconsistent implementations (for example: subdivision without creases, quad-only subdivision, etc.) are understood as “not standard” and temporary… which means OpenSubdiv needs to get it right!”

For more see



5 thoughts on “OpenSubdiv 3.0 coming (Alpha update)”

  1. You are right the MODO implementation is a branch of this technology not 100% the open source OpenSubdiv. You are quite right, but Modo was first to access the technology outside Pixar early on… and so deserved to be mentioned I thought.

    1. I see. Cheers.

      For anybody interested, Brad Peebler from The Foundry also wrote this about MODO’s implementation:

      “OpenSDS is great. However, our implementation solves topological changes MUCH faster than OSD. OSD is faster for deforming meshes, our DIRECT implementation is better for modeling. Are we watching OSD? You bet. We talk to Pixar all the time. I was just there last week meeting with Bill Polson. We’re on board the OSD train it’s just that we did it ourselves (with their assistance BTW) years before the OSD library existed. Will we add OSD library support in the future? Most likely but not as a replacement to what we have unless the topology changes are as fast with OSD as they are in our implementation.”


  2. I don’t think Arnold supports OpenSubdiv either. There were discussions on the mailing list but no commitment to support this format. You should check with Solid Angle to confirm this.

Comments are closed.