No-holds-barred Production Rendering

The perennial goal of creative professionals working on animation and special effects for motion pictures and television -- as well as computer-aided design (CAD) professionals -- is to deliver top-notch work on time and under budget.

While production times might be expected to decrease in direct proportion to available processing power, the ambitions of creative professionals are outpacing Moore’s law, which says that transistor density will double approximately every two years. Evermore complex rendering algorithms deliver increasing levels of visual subtlety and devour advances in processing power as quickly as they become available.

Look at some popular CG- and visual-effects-intensive films released over the last 10 years. The production rendering for Peter Jackson’s Lord of the Rings [1]  trilogy (2001, 2002, 2003) combined live action and highly realistic virtual characters. By the end of the trilogy, WETA [2], the visual effects firm, had upgraded its render farm to 3,200 processors. It reportedly took about four hours to render a single frame of Gollum [3] . Ratatouille [4] , released by Pixar Animation Studios [5]  in 2007, was rendered on an arsenal of 850 computer servers, which housed nearly 3,200 processors. Average rendering time: 6.5 hours. And James Cameron’s 2009 Avatar [6]  features animation and effects produced by WETA using a server farm of 4,000 blade servers hosting 40,000 processor cores. Rendering times ranged from 30 to 100 hours per frame.

In the design visualization world, no detail is too small to leave to the imagination, and the demand for high-quality visualization continues to grow in direct proportion to client expectations. The ability to render highly precise, photorealistic product designs and large-scale architectural plans, including exterior and interior walkthroughs, helps speed time to market. All these accomplishments are made possibly by multicore, multi-threaded processors, including Intel’s Xeon and Core i7 processors and AMD’s Opteron and Phenom processors.

Rendering at a Glance

Most popular rendering methods -- rasterization, ray tracing, ray casting and so on -- simulate light moving through a scene (also known as “light transport”) to produce photorealistic imagery or stylized non-photorealistic looks. At a basic level, renderers generally compute the visible geometry in a scene as triangles -- and paint them onto the virtual screen. Color and surface properties are computed by shaders, which commonly reference or sample texture maps.

Modern production rendering, however, is anything but basic -- motion blur, depth of field, deep shadows, subsurface scattering, color bleeding and caustics are just some of the myriad optical effects that can be simulated using one or more rendering methods.

Whether the goal is to achieve physical accuracy for things like medical imaging or product design, or to enhance reality for the sake of getting audiences to suspend disbelief as a story unfolds onscreen, the level of visual complexity is usually a balance between time, budget and hardware resources. Greater complexity usually equates with larger file sizes and heavier throughput computing workloads, because many rendering algorithms perform best when the entire scene file fits into physical memory, and main-memory-to-internal-memory transfers represent a critical performance bottleneck.

High-performance Processors Promote Speed, Complexity and Quality

 While artists and designers rarely have to think about the code that enables their digital content creation tools, they reap the rewards of well-crafted code optimized to take advantage of the breakthrough performance offered by both Intel Xeon processor-based and AMD Opteron processor-based workstations and servers. The ever-increasing capabilities and performance of multicore, multithreaded processing from both Intel and AMD provide an optimal foundation for a wealth of software innovations and rendering improvements for artists and designers, including these tools:

• Seriously Fast CPU-based Ray Tracing. The industry standard for ray tracing may be Pixar’s RenderMan, but Luxology [7] , developers of the award-winning Modo software, is no stranger to writing highly efficient code that’s optimized for maximum performance on Windows and Mac computers. The innovative 3D modeling and rendering software has been delivering fast ray-tracing-based rendering since 2006, when the Modo offline renderer and preview renderers were introduced.
• Efficient CPU-based Path Tracing. The Arnold Renderer by Solid Angle SL [8]  uses path tracing to achieve incredible realism. Path tracing is a type of ray tracing that is typically thought of as the most physically accurate rendering method, but it also has the reputation of being the slowest. Marcos Fajardo has been developing Arnold for 13 years, carefully tuning its algorithms for speed and performance suitable for demanding offline film production, which Sony Pictures Imageworks [9] used to render Cloudy With a Chance of Meatballs.
• Iterative Rendering on Modern CPUs. Visualization professionals need to take massively detailed CAD models from concept through construction with as much visual precision as time and budget will allow. Thanks to CAD applications such as Bentley Systems MicroStation [10] , visualization professionals -- engineers, architects, GIS professionals and owner operators -- can use the same ray tracing algorithms employed by Hollywood filmmakers.
• CPU-driven Interactive Rendering. KeyShot [11] is a CPU-driven interactive ray tracing and global illumination program developed by Luxion ApS for Intel-based Windows and Mac computers. KeyShot uses a physically correct and CIE-verified rendering engine featuring unbiased sampling techniques that compute mathematically correct results, scientifically accurate materials, multicore photon mapping, and adaptive material sampling. Alternately, if immediate gratification is your goal, there is Rendition [12] from Holomatix. Rather than producing an image block-by-block, Rendition constantly refines its output image, concentrating on areas that have the most detail first -- in real time.
• CPU-based Interactive Ray-traced Lighting. Maxwell Render [13] , from Next Limit Technologies of Madrid, is a ray tracer that utilizes unbiased algorithms to fully capture all light interactions between elements in a scene resulting in extremely realistic images. The software gives users a straightforward, easy-to-understand user interface and an effective workflow.

Computer-generated imagery is more sophisticated than ever. Real-world production rendering involves enormous scene files, multiple high-resolution texture maps and highly complex shader models to produce high-impact imagery with exacting realism -- a visual feat that would be impossible without multicore processors like the Intel Xeon and AMD Opteron processor families.