Intel has released a new paper that explains the most recent developments in path-traced light modelling and neural graphics research that its experts have been working on. One of the most significant advancements Intel is pursuing is considerably more efficient path-traced rendering, which might allow integrated GPUs to conduct path-tracing in real-time.
The post includes links to three new papers promoting new path-traced optimisations that the Intel Graphics Research Organisation will present at SIGGRAPH, EGSR, and HPG. These optimisations are intended to reduce the number of calculations required to simulate light bounces, hence alleviating and improving GPU performance.
The first paper displayed demonstrates a novel method for computing reflections on a GGX microfacet surface. GGX is a graphics technology that enables computers to collect simulated light bounces reflected in various directions. Materials are “reduced” to a hemispherical mirror that is far easier to replicate with this new technology.
The second study demonstrates a more efficient approach of creating glittering surfaces in 3D. Simulating glittering surfaces, according to Intel, is a “open challenge.”
With this new method, however, the average number of visible glitter from each pixel can be considered. As a result, the GPU only needs to output the appropriate amount of visible glitter to the eye.
Finally, another work, Markov Chain Mixture Models for Real-Time Direct Illumination, describes a more efficient way of creating optical trajectories in various illumination circumstances. The explanation is lengthy, but the final result is a more efficient rendering technique for real-time complicated direct illumination.
These three strategies will not, of course, guarantee that integrated GPUs can conduct path-tracing smoothly. However, they are geared at improving the key components of route tracing, such as ray tracing, shading, and sampling, which, according to Intel, will assist enhance real-time path tracing performance on integrated GPUs.
Also Read:
