VANCOUVER, B.C. —Computer simulations and video graphics are about to get a lot more real. Nvidia launched a new series of high-performance GPUs that greatly reduce the processing time for animation rendering from hours down to near real-time, according to company officials.
Company CEO and founder Jensen Huang took to the stage here at the annual ACM Siggraph conference to announce the new products before a packed hall of graphic artists and design professionals on Aug. 13.
The new Nvidia Turing GPU architecture uses machine learning, the company’s latest RT Core processors and Tensor Cores to make real time ray tracing possible. Ray tracing is a rendering technique for generating visual images designed to simulate real life objects, but it can often take many hours of computer processing time to complete the renderings.
“Our goal to create amazing imagery,” said Huang. “Turing opens up the $250 billion visual effects industry.”
While the initial target markets include movie and production studios, analyst Rob Enderle said because so many enterprises have video projects, servers based on Nvidia’s new architecture will find broad appeal.
“Nvidia is targeting rendering and photo realistic imaging, but you see a lot of enterprises doing these things beyond those in the movie business – for example, advertising and architecture design,” Enderle, principal analyst at The Enderle Group, told eWEEK
Huang noted that retailers such as IKEA render all images in its giant product catalog and it requires a server farm to handle all the processing. “Nvidia RTX brings accelerated workflows to this market,” said Huang.
A render farm of 240 Dual 12-core Skylake Intel CPU servers costs about $2 million, according to Huang, compared to a comparable setup of four Nvidia RTX 8-GPU servers that would cost $500,000, take up about a tenth of the space and require one-eleventh of the power to operate.
Because servers and workstations based on the new GPUs will speed up rendering, Enderle said it will be easier for companies to prototype consumer products such as cars and devices or all kinds without the same cost and time considerations.
“Now they’ll be able to do a lot of ‘what if’ new ideas and permutations of current designs in real time. No more waiting in the conference room for something to load or scheduling time to see it later,” he said.
Huang showed some simulated Star Wars footage the company had shown at its own developer conference in March that was created with the help of four Nvidia GPUs. “I have a surprise for you,” said Huang.
“What we just showed was a little faster and it was running on a single GPU,’ he said. “This is the world’s first ray tracing GPU, the Quadro RTX family that will come in several versions,” Huang said as he held up a Quadro RTX graphics card in his hand to show the crowd.
The Quadro RTX GPUs work with Nvidia’s Quadro Infinity software both of which are due to reach the market in the first quarter of 2019. Quadro Infinity lets multiple users access a single GPU through virtual workstations which the company said will “dramatically increase the density of the data center” saving both space and power.
The high-end Quadro RTX 8000 with 48GB memory is expected to sell for $10,000. Next is the Quadro RTX 6000 with 24GB of memory for $6,300. The Quadro RTX 500 with 16 GB of memory costs 2,300.
Enderle noted Nvidia’s GPUs are priced aggressively to seed the market as quickly as possible. He believes it’s all part of Nvidia’s broader mission to grab market share away from chip giant Intel. “The initial price is usually high, but clearly they feel they can justify a lower price because they expect to sell a lot of them,” said Enderle.
Huang said the new Turing processor contains a whopping 18.6 billion transistors and is the second largest processor in the world behind Nvidia’s Volta processor that is used in supercomputers. Nvidia’s previous Pascal processor has 11.8 billion transistors. Nvidia said applications will be able to simulate the physical world in virtual reality systems at 6x the speed of those running on the previous Pascal generation.
Major workstation and server vendors, including HPE, Dell and Lenovo, said they plan to use Nvidia’s latest chips in their systems.