Nvidia reveals new Turing GPU architecture, designed for raytracing and machine learning

Nvidia reveals new Turing GPU architecture, designed for raytracing and machine learning

Remember that Star Wars raytracing demo that was doing the rounds at GDC this year? It turns out that was done on one single GPU - the Quadro RTX.

This hardware supports 10 gigarays per second - faster than the most powerful GPU in the world at the moment. In addition, the tech is capable of 16 teraflops and 16 TIPS, as well as up to 500tn tensor operations per second for machine learning. Furthermore, with the new NVLink, the hardware can talk to other GPUs.

The tech can render raytracing 25 times faster than Nvidia's previous Pascal architecture.

"The new Turing GPU is the greatest leap since 2006 when we introduced CUDA," CEO Jen Hsun Huang said on-stage at SIGGRAPH.

"This fundamentally changes how computer graphics are going to be done. It's a step function in realism."

The assertion that the Star Wars demo was done on one GPU flies in the face of what was told at GDC this year by Nvidia - that it was being done via four of the firm's enterprise-designed Titan cards.

We can't wait to see the price of these new fancy cards being driven up by the cryptocurrency folks. 

Check out the full video below to see the man himself messing with the camera and going on about gigarays.

PCGamesInsider Contributing Editor

Alex Calvin is a freelance journalist who writes about the business of games. He started out at UK trade paper MCV in 2013 and left as deputy editor over three years later. In June 2017, he joined Steel Media as the editor for new site In October 2019 he left this full-time position at the company but still contributes to the site on a daily basis. He has also written for, VGC, Games London, The Observer/Guardian and Esquire UK.