What Does NVIDIA’s Ray Tracing News Mean for the CAD Market?18 Apr, 2018 By: Alex Herrera
Herrera on Hardware: At NVIDIA’s GPU Technology Conference in March, the graphics processing unit (GPU) developer announced “real-time ray tracing” — but is it? And what does this development portend?
Add ray tracing to that as one more compelling AI application for visualization professionals. Specifically though, what NVIDIA pitched at GTC ’18 was not just ray tracing — that technology is nothing new at all, but rather a high-quality rendering technique for photorealistic, physically based rendering we’ve used for years, whether we knew it or not. Rather, the company pitched something it claims is a first: ray tracing with real-time interactive performance, courtesy of the combination of a new ultra-high-end graphics card (the Quadro GV100) and new supporting AI-driven software (RTX technology). And any ray tracing that could be called real-time is not only new, but an achievement many computer graphics veterans would call a holy grail.
The first (and for now, the only) Volta-based Quadro GPU: the GV100. Image courtesy of NVIDIA.
Available immediately upon its GTC announcement, NVIDIA’s Quadro GV100 PCI Express add-in card pairs 15 TFLOPS of peak execution with 32 GB of max-performance HBM2 memory, providing a hardware foundation for raw graphics throughput beyond anything seen before in a single-card solution. But as much raw horsepower as the Quadro GV100 boasts, that alone wasn’t enough to get NVIDIA to its “first real-time ray tracer” goal. Rather, what closed the final gap to real-time levels was the company’s co-announced AI-driven RTX technology. RTX’s deep learning provided accelerations that proved just as instrumental in getting to real-time status as the usual bigger/better/faster semiconductor-driven advancements that come with any new-generation product. Specifically RTX, supported exclusively (for now, at least) on the Quadro GV100, leverages deep learning to recognize the eventual image as the ray tracer is still shooting and processing rays. RTX’s intelligence then fills in remaining pixels, de-noising the image and wrapping up the time-consuming rendering process that much quicker.
Think of this AI ray trace acceleration as 2D image recognition and filtering implemented over time, polishing each image without any temporal artifacts (which might emerge if each image was processed independently without taking into account how the previous and following image pixels were filled in). The end result? NVIDIA claimed 10X improvement over its previous GPU generation and that “first real-time ray tracing” title.
RTX technology built on top of the GV100 enables AI shortcuts in order to arrive at the desired ray-traced image much more quickly. Image courtesy of NVIDIA.
It’s worth pointing out that a claim to “real-time ray tracing” is a subjective one. I would define a real-time ray-tracing GPU as a single-card GPU solution that can produce images of significant complexity (i.e., number and type of objects, lights, and materials) at a quality level that the average viewer would say is comparable to a photo. And given that subjective but reasonable definition, this new Quadro GV100 with RTX certainly seems to fit the bill. And if scene complexity or quality goals get higher, two GV100s can be paired up via the card’s NVLink edge connector to ray trace that much faster.
A Brief Look at Ray Tracing
Using ray tracing to produce high-quality, photorealistic, computer-generated 3D imagery is a technique that’s been around a long time. Sift through proceedings from the SIGGRAPH conference (the epicenter of 3D graphics R&D in its heyday) published in the ‘80s and ‘90s, and you’ll see myriad papers dissecting and refining algorithms for ray tracing. Implementing physically based rendering (PBR), ray tracing by its definition emulates natural laws, producing images based on the precise physical properties and interaction of light, materials, colors, and the human visual response.
It traces a light ray in reverse as it traverses the scene — as it reflects, refracts, and is partially absorbed by surfaces until it eventually reaches a light source. By accumulating the contributions of a ray’s interactions with objects and lights, ray tracing achieves the ultimate in global illumination (accurate lighting and shading as a function of all lights, objects, and materials). That's something that traditional raster-based graphics can strive to approach, but often accomplish awkwardly and insufficiently.