The document discusses various techniques for realistic 3D rendering in real-time and non-real time contexts, including reflections, caustics, hair rendering, lighting and shadows. It compares the speed and quality of GPU real-time rendering to CPU rendering with and without baked textures or dynamic reflections. The document also discusses using render farms in the cloud to distribute rendering work for movies and games across thousands of CPUs and GPUs.
6. Hair rendering
● One of the most difficult
procedures
● simulation of naturally
looking hair and
movements
Hair rendering with D3D11 tessellation
NVIDEA GeForce Demo
7. Render Quality – Reflection example
3D Real-Time with 3D non-Realt-Time without 3D non-Real-Time with
Simulated Reflection Effects Reflection Effects Dynamic Reflections
9. Render farms for Rendering
● Connects thousands of CPU or GPUs to one cluster
● Amazon 1 cent per hour per virtual CPU
● Oracle/Sun 1$ per hour per real GPU
● Costs CPU = 1/100 GPU
10. Over-engineered?
● Big-Buck-Bunny (2008) by Blender
http://www.hpcwire.com/features/Rendering_in_the_Cloud_or_Not.html
Institute
● Rendered with Sun-Cloud
● 50k computing hours (5.7 years of
continuous rendering)
– Sponsered by sun for PR-campaign
● 1-2 hours of rendering time for a
single frame
● Total length of movie 11 min
11. Cloud rendering is a success story
http://www.tomshardware.com/reviews/render-farm-node,2340.html
● 40 million hours for Monsters vs. Aliens
● 30 million hours for Madagascar Escape
2 Africa
● 6.6 million hours for Revenge of the Sith
● 72 hours per single frame for
Transformers 2
12. Conclusion
● Simulating eyecandy effects in realtime rendering
is kind of the art
● Disadvantage: no dynamic “corrections” after baking
● Law of speed and costs: CPU ≈ 1 ⋅GPU
100
● Render farms “allow” for “battle of over-engineering”?
Tobias Guenther
Elaspix UG
Twitter: @elaspix