NVIDIA's Turing Tech Takes Real-Time Ray-Tracing GPUs to Next Level

At SIGGRAPH 2018, the graphics solutions developer celebrates achievement of a longtime goal, and launches a new graphics processing unit (GPU) lineup starting at $2,300.

You may know “Turing” as the name of a brilliant computer scientist — or of the eponymous test he developed, which determines whether a machine appears human during conversational exchanges. This week, NVIDIA introduced a new Turing to the world, in the form of a graphics processing unit (GPU) architecture that NVIDIA CEO Jensen Huang called the “greatest leap” in graphics progress since the invention of NVIDIA’s CUDA GPU, back in 2006.

At the same time, NVIDIA also revealed its first GPUs based on the Turing architecture. During his keynote presentation at SIGGRAPH 2018, Huang hailed the new Quadro RTX as “the world’s first ray-tracing GPU,” capable of rendering photorealistic scenes interactively, in real time. With the ability to render as many as 10 giga-rays (each representing one billion light paths) per second, “the performance [of the Quadro RTX] is absolutely incredible,” said Huang, who joked that it should be called “the world’s first giga-ray GPU.”

While graphics technology has become increasingly sophisticated with each passing year, the realistic simulation of light has remained a roadblock to creating photorealistic graphics in real time. Ray tracing, which calculates the paths taken by light bouncing off various surfaces and materials, convincingly mimics the behavior of light in the real world — but is very computationally demanding. Real-time ray tracing, in which users can interact with scenes without a delay for processing each time an element of the scene is adjusted, has been the Holy Grail of computer graphics, Huang said. Now, a “standard workstation” equipped with one of the new Quadro RTX GPUs is capable of real-time ray tracing, NVIDIA says.

If this announcement seems somehow familiar, you may recall that NVIDIA first announced real-time ray-tracing capabilities in March of this year, at its GPU Technology Conference (GTC). But as Alex Herrera pointed out in “What Does NVIDIA’s Ray Tracing News Mean for the CAD Market?” that initial step in NVIDIA’s plan did not represent “mainstream real-time ray tracing,” as the Quadro GV100’s price tag of $9,000 put it out of reach of the great majority of workstation users. The new Quadro RTX lineup, in contrast, includes three models: the 5000, 6000, and 8000 — priced at $2,300, $6,300, and $10,000, respectively.

Impact on Design Workflows

“We started this journey about 10 or so years ago … we thought we’d challenge ourselves to see if we could do it,” said Tony Tamasi, NVIDIA’s senior vice-president of content and technology. In addition to ray tracing, the company wanted to get users to switch from CPU-based rendering to a GPU-based option, he explained.

For CAD users, the real appeal of real time is the ability to see the ramifications of any design decision immediately. Without a need to wait or send files away from the local computer for rendering, they can instantly grasp the effects of changes — such as increased reflectivity when a finish material is updated, for example.

While a rough approximation of what something will look like is enough for some purposes, users in many fields — automotive design, AEC, film, and video games, to name a few — need photorealistic imagery. “Everybody is under pressure to deliver photorealistic images,” Huang said. “Architectural design needs photorealistic images … to convey the sense of the environment to their clients,” he observed, “Unless [the image] looks like [the final project] is going to look, it’s nearly impossible to use the image.” RTX technology is making it possible to bring GPU-accelerated workflows to these markets for the first time, he enthused.

The Turing GPU architecture incorporates two types of engines: ray-tracing RT cores, to accelerate ray tracing, and Tensor cores, for artificial intelligence (AI) inferencing. It is the combination of these two that enables real-time ray tracing. In addition to ray tracing, Turing also provides hardware acceleration of AI, programmable shading, and simulation.

RTXpansion Plan

In additional news, NVIDIA is also expanding its RTX development platform, which was announced in March. By adding new AI, ray-tracing, and simulation software development kits (SDKs), the company seeks to enable software developers to incorporate Turing functionality into their applications.

NVIDIA also unveiled Quadro RTX Server, a reference architecture for on-demand rendering in a datacenter. According to Huang, an RTX Server clocks in at one-quarter the cost of a typical $2 million render farm, takes up one-third the space, and consumes just one-eleventh the power. “Each megawatt of power [use equates to] a million dollars per year,” he noted.

In addition, the company shared the news that it has made its Material Definition Language (MDL) SDK, a high-level programming language, into an open-source offering. Developers can use MDL to apply realistic-looking materials to designs.

GPU Availability

Quadro RTX GPUs will become available directly from NVIDIA in the fourth quarter of 2018. They will also be shipping in workstations from original equipment manufacturers (OEMs), system builders, and distribution partners; ship dates will vary depending on the company. The 8000 features 48 GB of memory; the 6000 has 24 GB; and the 5000 has 16 GB.