At a new product extravaganza this morning during its GPU Technology Conference (GTC) in San Jose, Nvidia unveiled the new Volta-based Quadro GV100 with RTX technology, positioning the pairing as a transformative engine for real-time ray tracing, offering “the biggest advance in computer graphics since the introduction of programmable shaders 15 years ago.”
The Quadro GV100 GPU marks the introduction of the Volta architecture into the Quadro workstation accelerator line, and comes with 5,120 CUDA cores, 640 Tensor cores, and 32GB of HBM2 memory scalable to 64GB when you connect two Quadro GPUs using an NVLink bridge (as in the image below right). A single GV100 device delivers 7.4 teraflops at double-precision, 14.8 teraflops at single-precision and 118.5 teraflops tensor-based deep learning performance.
At the launch event this morning, Nvidia founder/CEO Jensen Huang discussed the complexity of image rendering saying the Volta-based Quadro GPU would enable workstations to take the place of supercomputers for high-demand rendering workloads.
As would be expected, Nvidia took potshots at rival Intel during a pre-announcement press briefing yesterday, declaring several times that Nvidia RTX delivers roughly 80x the rendering performance of CPUs.
“Nvidia has reinvented the workstation by taking ray tracing technology optimized for our Volta architecture, and marrying it with the highest performance hardware ever put in a workstation,” said Bob Pette, Nvidia VP of professional visualization. “Artists and designers can simulate and interact with their creations in ways never before possible, which will fundamentally change workflows across many industries.”
He acknowledged that CPUs are still dominant for rendering workloads.
“To be clear – because people didn’t quite believe that real-time rendering was achievable – most of the rendering today, to be honest, is CPU rendering,” Pette said. “But if you can get to a 10x, if you can get to a 100x, it is a game changer and we believe this will be a multi-million dollar market by the year 2020.” He added that “the industry impact of (the Quadro GV100) is going to be phenomenal given the amount of computing that industries spend on CPU cycles today…in a workflow that is now becoming way too long to make rapid (rendering) changes.”
Pette cited broadcasting as one of the industries.
“There are several billion hours of rendering time spent on hundreds of movies every year, more and more rendering is working its way into every movie, whether it’s improving the scene or improving the action in the movie,” he said. “You think about a CPU spending about 10 hours to render a frame that is one of 24 frames in a second, that is one of 60 seconds in a minute, that is one of ‘X’ number of minutes in a movie…”
In consumer markets, such as automobiles, Pette said enhanced rendering will enable consumers to quickly sample various colors and interior layouts, along with taking virtual test drives.
“Virtual reality (isn’t) just about building the car or the watch, it’s becoming more about selling it,” Pette said. “We all know as consumers – ‘I want to see it in this color, can I get it in this leather?’ The ability to do that in a split second and have it look just as real as in real life is what’s driving a lot of this demand for interactive realism.
“In the near future you’ll be picking your car, designing your car and driving your car in virtual reality. That’s got to be photo-realistic, that’s got to be a real road, that’s got to be real materials, in order for you to drive that car and buy that car without actually going for a test drive.”
Industry analyst Patrick Moorhead, president of Moor Insights & Strategy, said “ray tracing is one of the best ways to create completely realistic objects and scenes but has been hampered because it takes so much performance and bandwidth. Nvidia is addressing both these bottlenecks with the Quadro GV100. Nvidia is attacking the performance variable with Volta architecture compute and an AI denoiser and addressing bandwidth using the NVLink bus. This gets interesting for IBM POWER architectures as it’s the only architecture that supports NVLink [between the CPU and GPU].”
Pette said Nvidia RTX technology, introduced last week at the Game Developers Conference, is the result of 10 years of development work in computer graphics algorithms and GPU architectures. NVIDIA partnered with Microsoft to enable RTX support for applications that use Microsoft’s new DirectX Raytracing (DXR) API. He also said RTX is supported by more than 24 professional design and creative applications that have an aggregate user base of more than 25 million customers. Besides DXR, developers can access RTX through the Nvidia OptiX API, and, in the future, Vulkan, an open, cross-platform graphics standard.
Jacques Delacour, CEO of Optis, said the Quadro GV100 “accelerate(s) our simulation applications and algorithms, and Nvidia OptiX for fast, AI-based rendering. We’re excited about the potential Nvidia RTX ray tracing technology holds to deliver more lifelike images faster than ever.”
The Quadro GV100 is available now on nvidia.com for $8,900 (limit five at the time of writing), and, starting next month, from major workstation manufacturers, including Dell EMC, HPE, Lenovo and Fujitsu.