In Part 3: Basics of Ray Tracing, NVIDIA's Eric Haines runs through the basics of ray tracing hardware. The idea of making special-purpose hardware to accelerate ray tracing has been around for decades, with the AT&T Pixel Machine from 1987 being the oldest commercial effort. Progress since then has made even the lowliest cell phone today more powerful than that behemoth. However, Moore’s Law for CPUs has been slowing down, so more domain-specific solutions have begun to come to the fore. NVIDIA’s Turing architecture introduced acceleration for bounding volume hierarchy and neural net evaluation, both of which make real-time ray tracing feasible for more complex scenes. More details about this video can be found here: bit.ly/nvrttalk3
You can watch the all of the videos in the series on this KZfaq Playlist: • Ray Tracing Essentials
Ray tracing resources page: bit.ly/rtrtinfo
Ray Tracing Gems: raytracinggems.com
Additional links:
The initial image is by Alexia Rubod, and you can see more of her work here:
alexiarubod.com/#/new-gallery/
The “vast array of Crays” story is recounted here blogs.nvidia.com/blog/2018/08... by Turner Whitted
The sphereflake statistics are from this article www.realtimerendering.com/reso..., and the model generator is part of the free Standard Procedural Databases www.realtimerendering.com/reso... program set
You can try a program similar to the interactive ray tracer demo from 1987’s AT&T Pixel Machine on Shadertoy www.shadertoy.com/view/wdtSWf, it even works on your phone
The graph of Moore’s Law’s deceleration is after the graph in the article “A Domain-Specific Architecture for Deep Neural Networks.” dl.acm.org/citation.cfm?id=31...
The Star Wars image near the end is a still from the video Reflections, vimeo.com/291876490 by Epic, ILMxLAB, and NVIDIA
The interactive sphereflake footage at the end of this lesson actually shows 48 million spheres instead 5 million spheres, at a frame rate only somewhat slower