Questions about rendering are common to producers of CG content that is leveraged across multiple output media such as television, film and DVD. Let us assume for the sake of discussion that we are creating a hypothetical “backdoor pilot” of 75 minutes in length that may be released theatrically on film, broadcast on TV and ultimately end up on DVD. Let us also assume that we have settled upon an acceptable render time of 1 hour per frame, and have furthermore decided to work at a “film” frame rate of 24 frames per second.
FRAME RATE
Working at 24 fps is recommended for a number of reasons. In the first place, most animators are comfortable with it as a metric within which to work and to time movement (walk cycles, etc…). Working at 24 fps also tends to result in snappier animation than working at 30 fps (especially when “twos” are employed). While one might choose to animate at 30 fps (29.97 fps) for projects that are solely intended for video broadcast, 24 fps is generally preferred - even for projects with divergent final destinations. This is the case even when taking into account “3:2 pulldown”: the process of converting 24 fps material to 29.97 fps. While telecine conversion (which incorporates 3:2 pulldown) has been common to our industry for decades, it can result in troublesome visual artifacts – especially where lateral motion is concerned. And while there are ways to address these artifacts, that time and money is better spent elsewhere.
Simply put, the rule of thumb is: cater to your primary output medium, and adjust for the rest. For theatrical release, 24 fps is of course entirely compatible with the native frame rate of film, and also of digital projectors. Similarly for DVD, the MPEG-2 standard encodes source material at 24 frames per second. A flag is inserted within the MPEG-2 data stream that instructs conventional DVD players to perform a 3:2 pulldown in real-time (with the potential for artifacts). However, increasingly popular “progressive scan” DVD players react to this flag in a different way: creating high-quality, progressive video in real-time, with no degradation. These advantages should also translate to television as the Congressional mandate for 100% nationwide DTV broadcast takes effect on February 17, 2009.
FRAME RENDERING
There are two basic approaches to frame rendering: a global approach in which everything in the scene is rendered all at once, and a composite approach in which the scene is rendered in layers. Each approach has its pros & cons. The global approach treats the rendered scene like a live-action shoot in which everything is filmed at once, for better or worse. (Pixar has been one of the last notable adherents to this approach, but even they have turned to layered renders in recent films.) On the plus side, the global approach simplifies render organization and facilitates high-end, physically based global illumination. On the minus side, loading an entire scene into memory at once can result in a prohibitive footprint for complex environments. In addition, changes to any small part of a scene require that the entire environment be re-rendered. (In fact, the folks in Emeryville faced such prohibitively long renders on “Cars” that they ironically chose to address minor issues with digital “paint fixes” - just as one would on a live-action plate.)
The much more common layered approach distributes scenes to the renderer in layers defined by the artist and then assembled in a compositing package. While this approach requires careful organization, it also affords the ability to make targeted adjustments to specific elements without re-rendering the entire scene. Global illumination effects are either “faked” on the layers, or else are employed within specific layers (such as on Disney’s “Meet The Robinsons”, where global illumination was used in the background of certain scenes, with the characters composited atop). Either way, we should consider 1 hour of total render time per frame (on average) as a “reasonable” goal for an economical production – whether that hour is spent on one global render, or on a composite of layered renders.
RENDER FARMS
The fundamental components of a render farm are pretty simple: a collection of CPUs on a network for “cooking” the frames, a rendering application (such as RenderMan) to provide the “recipe”, a queue manager to distribute the scenes, and a network-accessible hard disk array for storing the data. As such, render farms can be created relatively cheaply from scratch, or purchased “pre-built” for a premium. But as with everything, the devil is in the details. Case in point: on “Cars”, Pixar saw their render times skyrocket to 10 hours per frame! And it turned out that the problem was not the CPUs themselves, but the NFS (“network file system”) server heads, which at only 1Gb of memory apiece were ill-equipped to handle the incredibly data-intensive scenes.
Moving to 32Gb server heads and replacing NFS with SAN (“storage area network”) - in which the devices appear to the operating system as locally attached - brought the “Cars” render times back down from 10 hours per frame to a much more reasonable 1 hour per frame. The moral of the story: not all render farms are created equal.
PUTTING IT ALL TOGETHER
So what does this mean to our hypothetical 75-minute backdoor pilot? At a frame rate of 24 fps, our 75-minute pilot amounts to a total of 108,000 frames. Assuming an average render time of 1 hour per frame, our project will take 108,000 hours to render: 4,500 CPU days, or a little more than 12 CPU years (the amount of time it would take one CPU to render the whole thing). At first glance, that may seem like a defeating number, but consider that 12 CPU years only takes four-and-a-half days to calculate on a 1000-CPU render farm, and a 3000-CPU render farm can render the entire project in only one-and-a-half days!
However, an animated film has never been made in which the entire project was forecast perfectly without preview and then sent off for a singular “first & final” render. 
Scenes are rendered and re-rendered many times in the troubleshooting of problems and in the pursuit of artistic quality. The more efficient this iteration is, the better. So truly, it is the WORKFLOW and not the hardware that makes or breaks a film.
And that is where production experience comes in.
