In 2018, NVIDIA released its RTX graphics cards based on its latest Turing microarchitecture. At the time, RTX was praised by industry insiders as being five years ahead of its time due to its real-time ray tracing and AI capabilities. Carl Flygare, product marketing manager for NVIDIA Quadro at PNY, considers RTX to be the biggest graphics advancement in over a decade.
“RTX really is the biggest performance and architectural disconnect since CUDA came out, all the way back in 2006,” he said.
That giant leap in hardware capability had one downside: software was still five years in the past. Independent software vendors (ISVs) had to scramble to bring the advantages of RTX into professional applications.
“The good news is that time is past,” Flygare said. “It's 2020. Developers are bringing applications to market that are RTX-enabled and support the RT Cores with real-time ray tracing, support the Tensor Cores with AI denoising or generative design, and support variable rate shading and foveated rendering in VR applications for product design.”
No longer shackled by software constraints, RTX is finally coming into play for engineers, architects and product designers. In this article, we’ll examine some of the ways that NVIDIA Quadro RTX graphics cards are being used to accelerate Industry 4.0 across four key stages of production: product design, advanced product design, design review, and post design.
Photorealistic Rendering
Let’s start with the titular benefit of Quadro RTX cards: real-time ray tracing (which is what RTX stands for). With Turing’s new RT (ray tracing) Cores, RTX cards provide hardware acceleration for ray tracing, the most photorealistic approach to rendering. The bottom line is that RTX enables better renders in less time.
Rendering is used throughout a product’s development, from renders of initial concepts and prototypes, to renders for design reviews, to the final renders used in product marketing. RTX cards aim to give engineers the ability to quickly produce photorealistic renders at any point in product development. To aid this goal, popular rendering applications such as SOLIDWORKS Visualize, Autodesk VRED, Siemens NX Ray Traced Studio, CATIA Live Rendering, V-Ray and more have all incorporated RTX technology.
AI and Machine Learning
The other headline feature of Quadro RTX cards is their built-in support for AI workflows. The cards use specially designed Tensor Cores—first introduced in the previous Volta microarchitecture, but improved in Turing—to accelerate the types of calculations used for machine learning. For example, NVIDIA uses Tensor Cores to assist in real-time ray tracing with a feature called AI-denoising.
“Even with RT cores working at up to 10 gigarays per second, ray tracing is an iterative process,” Flygare explained. “It needs a certain number of cycles to kick in before the scene is rendered with enough specificity to look good to the human eye.”
Without AI-denoising, ray tracing initially looks like a pointillist painting being generated dot by dot. But with AI-denoising, we don’t have to wait for each and every one of those dots to appear—machine learning completes the picture for us.
“After, say, 12 cycles of ray tracing, [AI-denoising] kicks in and fills in the blanks,” Flygare said. “And all of a sudden, you're looking at that shiny metallic paint on the complex body panel of the car under a canopy of trees on a sunny day. And it looks photorealistic, and it's done with a fraction of the compute performance in a fraction of the time a pure ray tracing approach would have required. So, AI-denoising can play a huge role with rendering.”
AI-denoising is just one of many ways that RTX cards can be used for artificial intelligence. “That one is so broad, it's almost easier to say what it won't touch,” Flygare joked.
Generative design, for instance, is an increasingly popular paradigm that uses algorithms to help product designers optimize their designs. Though these algorithms differ from case to case, machine learning could play a big role in the future of this technology. “Generative design is allowing engineers to optimize designs in ways that simply weren't possible before,” Flygare said.
Another example comes from the medical field, where medical devices could leverage RTX cards to analyze CT scans or MRIs before the data ever ends up in front of a radiologist. Security systems could use a similar approach to technologies like facial recognition. Consumer devices could likewise benefit in a multitude of ways.
“It's allowing your graphics board to become an assistant rather than something that spits out bits or does raw math at extremely high speeds,” Flygare said.
Real-Time Simulation
Another way NVIDIA Quadro RTX cards are contributing to Industry 4.0 is with their bread and butter, CUDA Cores. These are one of the main processing elements in NVIDIA Quadro cards, and RTX cards pack more CUDA Cores than ever before. It’s enough graphical computation power to enable a novel capability: real-time simulation.
For product designers, real-time simulation means no more waiting on analysts to send back their simulation results. Designers can conduct structural analyses, thermal analyses, fluid analyses and more in real-time, watching as changes to the model instantly propagate to the analysis.
The first software to make this possible was ANSYS Discovery Live, released with the Pascal generation of Quadro cards. It was Pascal that finally realized enough computing power for real-time simulation. “Until the raw FP32 performance of the Quadro card enabled that kind of capability, ANSYS couldn't build a product like this,” Flygare explained. “It was kind of like Chuck Yeager who broke the sound barrier; Quadro broke a compute barrier that enabled this kind of application to become a reality.”
With even more CUDA cores than the previous generation of NVIDIA Quadros, RTX cards are icing on the cake, in Flygare’s words. “With RTX, you've just got more headroom. So more challenging designs, more challenging simulations can be done in real time,” he said.
Virtual Reality
Virtual Reality (VR) is gaining popularity in product design, design review and post-design workflows, and NVIDIA Quadro RTX cards are helping it along. For one thing, RTX supports the recent VirtualLink connectivity standard, which requires only a single cable for head-mounted displays (HMDs).
RTX cards also support a feature called variable rate shading (VRS), which allows developers to customize the rendering quality of different areas of the screen. On its own, VRS could be considered a nifty trick for developers trying to get maximum performance from RTX cards. But the magic really happens when VRS is paired with a technique called foveated rendering.
Foveated rendering works with HMDs, like the HTC Vive Pro Eye, that use eye tracking to determine where users are looking. By taking this information and combining it with VRS, foveated rendering can ensure that wherever a user is looking, they’re seeing the highest quality render, and where they’re not looking, renders can relax. Since human peripheral vision naturally filters out details on the edge, its entirely natural to allow VRS to reduce the quality of these areas in VR. This frees up processing power and ultimately improves the VR experience.
“We now have application developers, like Autodesk VRED, that take full advantage of foveated rendering,” Flygare said. “And that helps designers or engineers make better informed decisions simply because they're seeing a higher resolution view.”
Virtualization
As product complexity grows, development timelines shrink, and engineers are working from more locations and on more devices, GPU virtualization is becoming increasingly important. And RTX is keeping pace. NVIDIA has introduced the RTX Server platform to enable desktop virtualization that can take full advantage of Quadro RTX cards.
RTX Server combines Quadro RTX cards (either the Quadro RTX 6000 or RTX 8000) with NVIDIA Quadro Virtual Data Center Workstation (Quadro vDWS) software. “NVIDIA RTX Server is a hardware reference platform and a software stack that has been developed by NVIDIA,” Flygare explained. “It's an open specification that system builders utilize to build RTX servers that can run a virtualization software stack and parse out rendering or other tasks to end users.”
NVIDIA Quadro RTX cards set a new milestone in NVIDIA’s support for desktop virtualization, and not just because of RTX Server. For the first time in Quadro history, NVIDIA offers the RTX 6000 and RTX 8000 cards with both active and passive cooling options. Active cooling is used in local workstations, but many server enclosures are designed for passively cooled components.
“NVIDIA's willingness to bring to market Quadro RTX products that are passively cooled is a first,” Flygare said. “They've never done that before for Quadro. So they're making it very clear that for GPU virtualization, the highest end Quadro boards available today are certified and supported by NVIDIA for data center deployment.”
RTX for Industry 4.0
With features for photorealistic rendering, machine learning, real-time simulation, VR, virtualization and more, NVIDIA Quadro RTX cards are ready for Industry 4.0.
“Buying RTX future proofs you, it gives you the capabilities you need for those applications that are coming out this year and will come next year and the year after that,” said Flygare. “It's really the right investment for engineers, designers and folks in manufacturing to be at the edge as software advances, to take advantage of the latest features in rendering or AI or virtual reality.”
For more information about RTX’s role in Industry 4.0, NVIDIA has published a series of whitepapers: Advanced Product Design For Industry 4.0.
To learn more about how NVIDIA RTX Servers can offer improvements over traditional CPU-based solutions, attend the PNY live webinar on April 23rd: NVIDIA RTX Server High-Performance Visual Computing in the Data Center