Graphics Reference
In-Depth Information
driven by the increasing need for interactive feedback in film projects. It has been
used in production at ILM. RenderAnt, a similar system developed by Kun Zhou
at Zhejiang University and his colleagues at Tsinghua University and Microsoft
Research Asia, presented in 2009, runs the entire Reyes
7
algorithm entirely in
hardware.
Another GPU-accelerated rendering system developed for the motion picture
industry is “Gelato” [Wexler et al. 05]. Released by NVIDIA in 2004, Gelato
was a hybrid GPU-CPU rendering system. At the time GPUs alone were not
sufficiently advanced to provide the high-quality results demanded by the movie
industry, so NVIDIA took the approach of leveraging the best features of CPU and
GPU computing. Although the development of Gelato was suspended in 2008, it
paved the way for important advances in GPU computing.
As GPU programming has increased in power and flexibility, GPU program-
ming environments have become much more user friendly. This has led to the idea
of using GPUs for a wide range of tasks beyond rendering, a philosophy that has
come to be called
general purpose GPU
(GPGPU) computing. In line with this
philosophy, NVIDIA released a new GPU programming language in 2006 named
CUDA
(from
compute unified device architecture
). CUDA is a C-like language
for parallel computing using a shared-memory model. A primary advantage of
CUDA over direct GPU programming is the shared random-access memory (re-
call that ray tracing and photon mapping on the GPU originally required data
to be encoded in texture maps and other buffers). The release of CUDA has
made it possible to exploit the power of graphics hardware for general-purpose
programming.
Since 2008, the use of ray tracing has become more and more prevalent in
the entertainment industry and in the CG industry in general. In 2009 NVIDIA
announced plans to release a GPU ray-tracing engine known as OptiX. At its
heart, OptiX is a ray-tracing library for CUDA. It has the appeal of lowering the
development costs of producing high-performance ray-tracing software. It is still
in the early stages at the time of this writing, but expectations are high and the
hope is that it will contribute to widespread use of ray tracing in the future.
Since the emergence of commercial programmable graphics hardware around
the year 2000, research into GPU rendering has become a major trend in the CG
industry. Much of the work has gone into implementing existing algorithms on
GPUs. Now that this has largely been done, research is being aimed at what
is really needed from graphics hardware. The field of GPU computing is still
new and rapidly evolving. The industry is anxiously looking forward to future
developments.
7
Reyes is an algorithm and rendering system developed at Pixar for rendering RenderMan images.
