Latency is among the biggest challenges in computer-generated animation. Graphic artists have long been able to make quick changes to hair, clothes, plants and other details, but they’ve generally had to wait minutes—or longer—to see the finished result.
But as GPU technology has matured, that excruciating wait-time has all but disappeared, Laurence Emms, a technical director at Pixar, told a packed session at the GPU Technology Conference.
This is especially true in the early stages of an animated film, when GPUs are relied on to do the heavy lifting of rendering animated characters and backgrounds. But Pixar is pushing the use of GPUs later and later in the process, enabling their use for more detailed tasks, such as perfecting hair.
“It’s about speed, and letting the artists see what they’re doing quicker,” Emms said after the session.
For instance, Pixar uses tessellation shaders running on GPUs to preview characters’ hair styles. Among the things they’re looking for are stray hairs, consistency of thickness, and whether hair will block something important behind it in the final shot. Any such rendering on GPUs also must match renders on the CPU, which is where Pixar runs its proprietary RenderMan software.
Meanwhile, a new type of shaders—geometry shaders—has “opened up a new era of GPU computing at Pixar,” says Emms, helping in particular with rendering vegetation in real time.
To illustrate, Emms shared an animation demo of simple geometric cubes falling on top of each other and reacting to the resulting force, showing how a mass spring simulation is rendered on CUDA-enabled GPUs.
“Everything, including collision detection, is running on the GPU,” he said.
Recent advances in GPU technology are pushing Pixar’s capabilities even further. Emms said the studio is excited about the potential for the dynamic parallelism of NVIDIA’s Kepler architecture to speed up more processes, allowing Pixar to follow-through on its plans to rely on GPUs further into the moviemaking process.
And recently, Pixar has open sourced its GPU-powered OpenSubdiv surface rendering libraries, in the hope that other technology partners can incorporate its innovation into other film-making tools.
