Intel reports high-rendered graphics with low-power GPU

This article has been reviewed according to Science X’s editorial policies and procedures. The editor has highlighted the following attributes while ensuring the credibility of the content:

Reality check

pre-prints

reliable source

proofreading






We introduce a new importance sampling algorithm for GGX microfacial BSDFs. Our algorithm provides the same variance reduction as the most advanced method [Hei18] but with lower computational cost. This translates into a slight but systematic increase in efficiency to show e.g. raw conductors (left) and raw dielectrics (right). Scene credits: Yasutoshi Mori for the PBRT-v4 renderer. Credit: arXiv (2023). DOI: 10.48550/arxiv.2306.05044

Aiming to develop graphics processing units in a burgeoning AI universe, Intel has announced the release of several papers outlining the efforts it is pursuing in what observers perceive as billion-dollar opportunity in the coming years for the semiconductor chip giant.

Intel is presenting seven papers in three conferences on advances in computer graphics.

The first papers were formally presented at a joint conference last month organized by the High Performance Graphics (HPG) forum and the Eurographics Symposium on Rendering at Delft University of Technology in the Netherlands. The remaining papers will be discussed at a conference organized by SIGGRAPH (Special Interest Group on Computer Graphics and Interactive Engineering) in August.

The main focus is on how to improve historically heavy rendering workflows.

The articles specifically discuss two processes, ray tracing and line tracing. Both are used for realistic image reproduction, especially in gaming, where accurate representation of the physics of light is crucial for natural looking images.

Ray tracing applies algorithms to track the trajectory of light waves and calculate hue, reflection, and shadow values. The enormous processing power required for real-time rendering is so great that the frame rate often has a noticeable impact.

Path tracking may require heavier processing. It follows multiple light rays, tracks paths as they reflect off surfaces, and interacts with light among other elements. A process known as Monte Carlo integration helps to determine precise shade and color values.

Intel says these tracking methods can be made more efficient. One of its papers, «Visible GGX Standard Sampling with Spherical Exponents», describes an innovative approach to computing completed hemispherical items that «accelerates the system in benchmarks» ours».

Another article showed a 500% improvement in speed when rendering «sparkling» objects like speckled car paint, snow, molded plastic, and flowing water. «Real-time rendering of sparkling interfaces using distributed binomial laws on anisotropic grids» explains that current methods achieve stunning realism but «have a very high cost». high» in terms of processing power and speed.

In a paper that will be discussed at the SIGGRAPH conference in August, Intel will look at advances in neural graphics, an approach the company says is «revolutionizing the graphics field.» It is used to quickly scale high-quality graphics on games and movies.

«The new level of neural detail representation achieves 70%–95% compression compared to ‘vanilla’ path tracking,» Intel reports.

Other papers examine improvements in the representation of translucent materials and «sampling photon trajectories in difficult lighting situations.»

Ultimately, Intel hopes significant advancements in processing methods will allow users to enjoy real-time visuals without the need for a high-powered GPU.

«The new building blocks presented at this year’s conference, coupled with our broad offering of GPU products and our scalable cross-architecture rendering stack, will help developers and enterprise to perform more efficient rendering of digital twins, as well as future realistic AR and VR experiences as aggregate data for sim2real AI training,» Intel stated in its blog.

Intel plans to make its work open source.

Intel articles appeared throughout June on the preprint server arXiv.

More information:
Jonathan Dupuy et al, Visible GGX Norm Sampling with Spherical Hats, arXiv (2023). DOI: 10.48550/arxiv.2306.05044

Real-time rendering of sparkling interfaces using distributed binomial laws over anisotropic meshes, arXiv (2023). DOI: 10.48550/arxiv.2306.05051

Laurent Belcour et al., Upsampling one-to-many spectroscopy of reflection and transmission, arXiv (2023). DOI: 10.48550/arxiv.2306.11464

Intel Blog: www.intel.com/content/www/us/e… ative-ai-update.html

Journal information:
arXiv

#Intel #reports #highrendered #graphics #lowpower #GPU

Deja un comentario