Four animated films starring Shrek, the world’s most lovable ogre, have graced the silver screen over the last decade, each achieving significant breakthroughs in computer-generated imagery and each building upon the latest advances in processor technology.
The final film in the franchise was released in May 2010, making this a good time to consider what DreamWorks Animation has accomplished over the course of telling the Shrek saga. The studio has used every technique, tool, and processor cycle at its disposal to elevate the believability, expressiveness, and pure animated impact of the characters in this iconic film series.
Processing Power Drives Advances in Animation
One way to look at the advances in animation is through the virtual eyes of Shrek, a character first introduced by DreamWorks Animation in 2001. Shrek won the first Academy Award for Best Animated Feature and established DreamWorks Animation as a leader in CG animation.
Over the years and throughout the production of the Shrek franchise, a number of DreamWorks Animation’s technological advances have been enabled by Intel® technology.
As DreamWorks Animation CTO Ed Leonard said, “We talk about render power and the visual imagery that it can provide. We talk about processing cycles as a core filmmaking aspect as much as we do storyboards and cast. To say that without processing power we wouldn’t have a film is an understatement. It is among the essential elements in bringing great films to life.”
Render Hours on the Rise: Shrek’s Law Meets Moore’s Law
Since 2001, many of the advances in producing animated films have been forged by leaps in processing power. Comparing the basic metrics of the animation process (time to render frames, availability of special effects, real-time visualization, complexity of scenes) provides a startling indication of just how far this technology has progressed in less than 10 years, as shown in Figure 1, which charts processing power needs for Shrek through Shrek Forever After.
DreamWorks Animation coined “Shrek’s Law,” a tongue-in cheek corollary to Moore’s Law. According to Shrek’s Law, the CPU render hours needed to complete production on a theatrical sequel will double compared to the amount of time needed on the previous film. This law has proven itself true on the Shrek movies: In 2001, Shrek required approximately 5 million CPU render hours. In 2004, Shrek 2 required over 10 million CPU render hours. In 2007, Shrek the Third required over 20 million CPU render hours. Finally, the 2010 3D release of Shrek Forever After topped the charts at more than 50 million CPU render hours. All of this was achieved without significantly expanding the size of DreamWorks Animation’s data center.
While delivering the keynote address at CES 2010, Paul Otellini observed, “As you know, Intel is a company built on Moore’s Law, a law that postulates we can double the number of transistors on a chip every 24 months. The first microprocessor that was in that film clip [shown on-screen] was the 4004. It had a grand total of 2,251 transistors. The new processors that we’re introducing at CES today have over a billion transistors.”
“What I find interesting about Moore’s Law, though,” Otellini continued, “is that it’s not a law of nature. It’s a law that reflects human inventiveness. We’ve been able to advance technology with consistent predictability by embracing the unpredictability of the discovery process. Every two years, we schedule a breakthrough. Each step along the way takes six years to build and costs [USD] 12 billion a step. That’s an expensive proposition, but it’s one that has enabled the products that have fueled the computer industry for over 30 years.”
Putting Additional Processing Power to Work
At DreamWorks Animation, this geometric increase in compute power has enabled a vast range of new visual capabilities. Let’s take “global illumination” as an example. As Lincoln Wallen, head of research and development at DreamWorks Animation, explained, “Global illumination is the ability to have everything interact, in terms of light.”
Consider a character standing next to a red wall. The wall itself is not a true light source but will reflect light that strikes its surface and create a slightly red glow. This affects the lighting on any character standing next to that wall. This kind of bounce lighting is the essential idea behind global illumination.
Processing power helps in terms of all the interactivity—being able to see things quickly and keep the creative flow. You don’t want the artist waiting for the computer to react. You want artists to ask their computers to do something and for those things to happen as quickly as possible so that they can go on to the next thing. Waiting doesn’t help the artistic process.”
—Derek Chan, Head of Digital Operations, DreamWorks Animation
“Global illumination,” Wallen continued, “is a very complex factor to calculate, because you have to not only cast rays of light, but also calculate the bounce and angle of those rays. It certainly has a notable impact in terms of the fullness of the color and, again, it is hard to see specifically other than if we turned it on and then turned it off, you would say, ‘Oh, that really fills in a lot of the detail.’ Global illumination creates a lot more ambient light.”
This subtle but important visual effect wasn’t technologically possible during feature film productions that preceded 2010’s Shrek Forever After. From a processing standpoint, during production on the original Shrek, global illumination was too expensive computationally. In comparison, Shrek Forever After takes full advantage of the added realism of global illumination. Touches like this may be subtle but they cumulatively have a large impact on the visual experience of DreamWorks Animation’s films.
Simulating the Behavior of Everyday Objects
Simulating the behavior of everyday objects in our surroundings—the flow of water, the effects of wind, the movement of fabrics worn by a character—adds a substantial degree of processor intensive work to the animation pipeline. For example, creating realistic hair on characters is a task that quickly consumes the available processing power.
“Hair is complex in terms of simulation.” Wallen said. “Fiona’s hair in the original Shrek was styled as a braid so that it could be treated as a single object. At that time, you didn’t have characters with long, flowing hair in the film because from a simulation standpoint it was very costly to try to do.”
By Shrek Forever After, processing power allowed DreamWorks Animation artists to unbraid Fiona’s hair. In one scene, she takes off her helmet with the wind blowing fiercely, and her long hair flows in the wind—providing a great creative moment that speaks to her character, one that simply couldn’t have been achieved just a few years earlier.
Animators also rely on additional processing power to provide quick feedback on animated movements, to handle larger numbers of control points in animated characters, and to provide pre-visualization of various effects. As Derek Chan, head of digital operations, said, “Processing power helps in terms of all the interactivity—being able to see things quickly and keep the creative flow. You don’t want the artist waiting for the computer to react. You want artists to ask their computers to do something and for those things to happen as quickly as possible so that they can go on to the next thing. Waiting doesn’t help the artistic process.”
Mini-Farms Offload Tasks to Boost Responsiveness
The use of mini-farms—small clusters of multi-core processor systems—was introduced in Shrek the Third to offload the processing of complex tasks. It was then used more extensively in Shrek Forever After. Chan explained that during the production of the original Shrek, artists worked with a computer on their desks, typically a single machine with one or two cores. Each small change that was made, such as moving a light or posing a model, required the computer to go through a lengthy recalculation cycle before the artist could gauge whether the effect or movement worked in the scene.
“In Shrek the Third—and certainly in Shrek Forever After— instead of having one computer doing the commands, we had a mini-farm,” Chan said. “The mini-farm was a set of machines, each with multiple cores, in the data center. The machines were dedicated to accelerating the work the artists were doing on the desktop. So, the scheduler would tile your shot or your image into many more tiles than there are cores and farm those jobs to the mini-farm machines for processing. Each of those machines had to compute only a tiny section and then send that result back. By giving the artists ten or twenty times the processing power at the desktop, we could significantly accelerate those tasks—which used to take hours to perform—down to minutes or seconds.”
Parallelism, Chan observed, is the key to many of these performance gains. “With the mini-farm,” Chan explained, “the artist makes the necessary changes, clicks ‘recompute,’ and, whoosh, it is back in seconds or minutes, not hours!”
The Power of a True Stereoscopic 3D Experience
Shrek Forever After is an especially noteworthy accomplishment because it was authored in stereoscopic 3D. This affected everything, not just the character design and movement, but even the backgrounds.
“A 2D matte painting doesn’t work well in 3D,” Chan said, “because your mind wants to see the geometry in the background from two different perspectives. This causes some challenges when you are used to using images to represent things that would then look flat in a 3D world.”
Chan points out that this forces a lot more geometry to come into play. If you see a tree in the background in a stereoscopic 3D film, your mind expects to see some depth. Is that branch coming toward me? Are those leaves closer to me? Are they moving appropriately? Although this is something that most people wouldn’t naturally think about, it’s a very noticeable part of the stereoscopic 3D viewing experience.
Chan continued, “What we want is a viewing experience where you forget that you are seeing it in 3D, where your mind just gets used to the fact that it is seeing in 3D and it is not noticeable other than: Wow, this just feels more rich and involving and immersive.”
— Puss in Boots
DreamWorks Animation Innovates with Cloud Computing
The future of CG rendering is definitely in the clouds. During the production of Shrek Forever After and How to Train Your Dragon, DreamWorks Animation pushed the state-of-the-art and used cloud computing to complete over five million render hours in a New Mexico-based compute facility. Addressing peak demand with a utility service model is highly scalable and cost effective.
“Our success with both How to Train Your Dragon and Shrek Forever After is working proof that high-performance computing clouds are not only possible but practical,” said Ed Leonard, DreamWorks Animation’s chief technology officer. “Having the ability to instantly scale compute resources to meet the demands of our industry-leading artists is one of the holy grails in CG rendering. There is no doubt in my mind that this is the future of rendering.”
This model of providing cloud computing services to deliver massive amounts of computational resources on demand promises to change the approach used in a variety of compute intensive activities.
----------------------------------------
Sign up for Intel®Visual Adrenaline magazine: http://va.softwaredispatch.intel.com/
