When real-time ray tracing was first announced in 2018 with Nvidia's 20-series Turing GPUs, it was nowhere near ready for prime time. Only a handful of games supported it, GPUs could barely run it at 60fps, and AMD had nothing to offer in response. Now that the technology has matured over three generations of GPUs, Nvidia recently updated it with path tracing, also known as "full tray tracing," as it can take infinite light sources into account. Though it looks fantastic, the downside is it can only run at 12fps on an RTX 4090 without upscaling. Now researchers from Intel are proposing several novel methods they say can make path tracing more efficient, hoping to allow it to run on lower-end GPUs in the future. It will present its research results at the upcoming SIGGRAPH conference.
Intel presented some of its findings in a recent blog post and noted it's already shared some of its research papers at recent industry events for graphics professionals. It describes its interest in path tracing as being related to its "ubiquitous" integrated graphics and its "emerging" Arc GPUs for clients, professionals, and data centers. With Intel already entrenched on the iGPU side of the equation and hoping to gain a foothold in the discrete market, it's trying to make path tracing more efficient so its GPUs can take part in this revolution of photorealism, essentially. Path tracing has to be made more efficient, and Intel is proposing several novel methods for speeding up the technology. Intel's blog post was flagged by Tom's Hardware.
The first paper Intel published deals with increasing the speed of shading calculations. This is the process whereby a light source interacts with an object, and how the light's reaction has to be simulated according to a distribution model based on the kind of material involved (soft, hard, textured, etc.). The first discovery is designed to speed up calculations of light reflecting off a GGX microfacet surface, the default surface type used for simulations in the graphics industry. Intel's research shows that changing the model's surface to a hemispherical sphere and a new algorithm to calculate reflections can significantly speed things up. In the two examples provided by the company, there's an increase in rendering time of almost 4% and 8%.
The second topic is rendering "glinty" objects, usually artificial objects that appear shimmering. Intel says it's introducing a "statistical law" regarding the average number of glints visible by the camera or player at any given time. This approach reduces the number of calculations involved, as only those visible at any given time are rendered. Though Intel says this can be run in real-time on desktop PCs, simulating these surfaces in video games is still an "open challenge."
Since ray (and path) tracing is all about calculating light reflections, Intel says these new approaches allow for more efficient path tracing in the future, which could theoretically allow it to be run on midrange GPUs and eventually even integrated graphics. Intel says it will make this research available as an open-source project for developers and practitioners, so perhaps AMD and Nvidia will even take a peek at it.
