Nvidia Shines New Light on Graphics Processing

If a picture is worth a thousand words, can cinema-quality image rendering for video games be worth $10,000?

That’s a question soon to be answered with the launch of Nvidia’s latest graphics processing unit (GPU), a purpose-built architecture designed to deliver ray tracingOpens a new window (RT) technology for more lifelike rendering of images in 3D environments.

The company unveiled its Quadro RTX GPUs at last week’s SIGGRAPH 2018 conference for design professionals in Vancouver, Canada. At the Gamescom event that opened August 21 in Cologne, Germany, attendees enjoyed a preview of the same technology to be deployed in an updated family of GeForce graphics cards for gaming computers.

Simulated Dimensions

RT involves generating paths of light as pixels, tracking those rays as they travel through a given scene and simulating the reflection, refraction, scattering and dispersion that occurs when light rays encounter geometric objects to simulate the look and feel of 3D. The technology expedites a resource-intensive process used in the production of games and films.

Instead of the hard lines between light and shadow that occur when cube mapsOpens a new window render graphic images, Nvidia’s RTX technology uses algorithms and neural networks to replicate the physics of light rays as they meet solid, transparent and translucent objects. And in real time.

An artificial intelligence engine and graphics accelerator combine in what Nvidia has dubbed its “Turing architecture.” Named for Alan Turing, the British logician credited with breaking the Enigma code used by Nazi Germany to communicate orders and troop movements during World War II, the architecture uses RT Cores and Tensor Cores in parallel to process and render rasterized images on-screen.

Faster Processing

The result permits off-screen light sources to illuminate and reflect visible objects as gamers move through on-screen environments. The GPUs eliminate the need for cube maps (which lengthens the time to render images by requiring images be processed six times to lessen distortion).

Enter Turing architecture for the execution of this high degree of real-time computation. Drawing on the power of neural networks enables RT to occur as much as 25 times faster than the GPU architecture, called Pascal and introduced two years ago, that it replaces. For final-frame rendering of filmed effects, the Turing-enabled GPUs can execute those processes 30 times faster than with an ordinary CPU.

Single Chip Tech

Nvidia unveiled its real-time RT technology in March at the GDC conference for game developers in San Francisco. As then demonstrated, RTX operated through a pair of Tesla V100 GPUs that permit the execution of the artificial intelligence training and workloads to accurately fill pixel gaps in light and shadow at speed.

Given parallel processing capabilities that reach to 16 trillion calculations per second, the Turing-class GPUs make possible RT, AI, programmable shading and simulation – and all via a graphics card with a single chipset.

The Quadro RTX GPU puts 4608 CUDA parallel processing cores and a neural network comprised of 576 Tensor Cores in the hands of design professionals, enabling them to draw on 48 GBs of memory to render graphics at 10 Gigarays (10 million light rays) per second. Deployed in combination via Nvidia’s NVLink 2 construction, multiple Turing-class GPUs permit faster production of light and sound effects for film production.

For gamers, Nvidia’s GeForce 2000 series delivers the same rates of Gigarays at performance levels six times greater than the company’s GeForce 1000 series graphics cards that are among the world’s bestsellers.

Tools Included

Working with Microsoft, Nvidia incorporated RT technology into the operating system giant’s DirectX API development platform. To further assist developers, Nvidia offers a toolkit for the software to run on the Turing GPU architecture.

A deep-learning NGX stack enables them to modify existing applications for greater performance and ease of use.

A host of partner/developersOpens a new window have signed on to provide RTX support for their graphics programs, among them Adobe, Autodesk, Octane Render, Redshift and Siemens NX. Together, they supply many of the applications used by the global design community.

CEO Jensen Huang said the RTX as enabled by the Turing GPU achieves the “Holy Grail” of image rendering for films and games. He added the RT technology a decade in development enabled Nvidia to beat industry expectations that real-time rendering was five years off.

Mobile in the Works

While mobile applications remain in the future due to the physical size of the Turing GPUs, Nvidia is touting the value of Quadro RTX deployment in its datacenter business, where strong growthOpens a new window is underpinning the company’s recently released quarterly results.

As such, time will tell whether prices from $2,300 to $10,000 for those units when they hit the market in Q4 will be too much for design and movie professionals. Judging by the response from attendees at Gamescom, the GeForce units that will appear next month priced from $600 to $1,200 should be a faster sell among gamers eager for a look at what the company is heralding as “the next industrial revolution.”