Chipmaker Nvidia Corp. late Monday unveiled the next generation of its graphics processing technology based on its long-awaited Turing architecture.

The announcement represents a major upgrade from its Pascal chips that were introduced in 2016. Nvidia Chief Executive Jensen Huang (pictured) introduced the new chips and architecture during the annual SIGGRAPH computer graphics event in Vancouver, announcing them as the “world’s first ray-tracing GPU.”

The new family of graphics processing units, called the Quadro RTX, performs at up to “10 gigarays per second” with “16 teraflops of single-precision performance.”

Huang said the Turing architecture is the “Holy Grail of computer graphics” and its most important innovation in more than a decade, and the greatest leap since the invention of the CUDA GPU in 2006.

Turing replaces Nvidia’s previous architecture Pascal and ushers in a number of important advances, the most important being “ray tracing,” which is a technique for rendering graphics far more realistically than its older technology could ever achieve.

Ray tracing enables the superrealistic rendering of light and shadows in a computer-generated scene. It uses an algorithm that can trace the path of light and then simulate the way the light interacts with the virtual objects it falls upon in the computer-generated world.

Up until now however the process has been strictly reserved for major film making studios as only those with the biggest of budgets could access the massive computer processing power needed. But thanks to its new architecture and chips, Nvidia is making the technology much more widely available.

The Turing architecture’s ray-tracing engine is powered by new “RT Cores” that are used to accelerate computing operations to determine how light can interact with its environment in a 3-D space. Nvidia reckons its new Quadro RTX chips can boost ray-tracing performance by up to 25 times that of its Pascal chips, performing the necessary calculations at speeds of up to 10 gigarays per second.

Nvidia first showed off its new GPU’s ray-tracing capabilities during the company’s GPU Technology Conference in March through a computer generated imagery Star Wars clip created especially for the demonstration.

The Quadro RTX GPUs will launch in the fourth quarter in three versions: the Quadro RTX 8000, the Quadro RTX 6000 and the Quadro RTX 5000.

The GPUs also feature something called “Tensor Cores,” which are processors that can accelerate machine learning training and inferencing, Nvidia said. The Tensor Cores will provide speeds of up to 500 trillion tensor operations per second.

Ray-tracing performance can be boosted even further by pairing the GPUs with Nvidia’s NVLink technology, providing up to 96 gigabits of memory and 100GB-per-second data transfer speeds from two RTX 8000 chips.

Nvidia said this kind of performance would enable the creation of artificial intelligence-powered applications with new capabilities such as deep learning antialiasing, which is a new kind of motion image generation technology. Other applications include “denoising,” resolution scaling and video retiming.

“This is truly a breakthrough and five years ahead of schedule,” Jon Peddie, president of Jon Peddie Research, told SiliconANGLE. “The combination of Nvidia’s hardware accelerators and software development have made a quantum jump. Ray tracing has been the Holy Grail of computer graphics for decades, and once it gets fully understood and investigated, all kinds of new applications will emerge.”

As to what specific kinds of applications we might see getting built off Nvidia’s new GPUs, Peddie said the impact would be widespread, covering everything from acoustics to medical diagnosis and even astrophysics. Ray tracing also has great potential in machine learning applications, particularly when it comes to sharing workloads or doing things serially such as computing and then rendering, where the RTX series would provide “unimaginable ROI,” according to the analyst.

Nvidia is also making available its new RTX graphics platform for developers to make the most of the new chips. The platform includes application programming interfaces for ray tracing, AI, rasterization and simulations. The platform also supports Nvidia’s new open-source Material Definition Language software development kit, which is a suite of tools that integrate the precise look and feel of real-world materials into rendering applications.

Nvidia said the RTX 8000 chips come with 48GB of memory and will go on sale at an “estimated street price” of $10,000. The RTX 6000 with 24GB of memory will cost $6,300, while the RTX 5000 with 16GB of memory is priced at $2,300.

Image: Nvidia