Home / Tech News / Nvidia’s new Turing architecture is all about real-time ray tracing and AI

Nvidia’s new Turing architecture is all about real-time ray tracing and AI


In current days, phrase about Nvidia’s new Turing structure began leaking out of the Santa Clara-based firm’s headquarters. So it didn’t come as a serious shock that the corporate at present introduced throughout its Siggraph keynote the launch of this new structure and three new pro-oriented workstation graphics playing cards in its Quadro family.

Nvidia describes the brand new Turing structure as “the best leap for the reason that invention of the CUDA GPU in 2006.” That’s a excessive bar to clear, however there could also be a kernel of reality right here. These new Quadro RTx chips are the primary to function the corporate’s new RT Cores. “RT” right here stands for ray tracing, a rendering methodology that mainly traces the trail of sunshine because it interacts with the objects in a scene. This method has been round for a really very long time (keep in mind POV-Ray on the Amiga?). Historically, although, it was at all times very computationally intensive, although the outcomes are likely to look way more life like. In recent times, ray tracing acquired a brand new increase due to quicker GPUs and assist from the likes of Microsoft, which just lately added ray tracing assist to DirectX.

“Hybrid rendering will change the trade, opening up superb potentialities that improve our lives with extra stunning designs, richer leisure and extra interactive experiences,” stated Nvidia CEO Jensen Huang. “The arrival of real-time ray tracing is the Holy Grail of our trade.”

The brand new RT cores can speed up ray tracing by as much as 25 occasions in comparison with Nvidia’s Pascal structure, and Nvidia claims 10 GigaRays a second for the utmost efficiency.

Unsurprisingly, the three new Turing-based Quadro GPUs will even function the corporate’s AI-centric Tensor Cores, in addition to four,608 CUDA cores that may ship as much as 16 trillion floating level operations in parallel with 16 trillion integer operations per second. The chips function GDDR6 reminiscence to expedite issues, and assist Nvidia’s NVLink expertise to scale up reminiscence capability to as much as 96GB and 100GB/s of bandwidth.

The AI half right here is extra vital than it might appear at first. With NGX, Nvidia at present additionally launched a brand new platform that goals to deliver AI into the graphics pipelines. “NGX expertise brings capabilities corresponding to taking a regular digicam feed and creating tremendous sluggish movement such as you’d get from a $100,000+ specialised digicam,” the corporate explains, and likewise notes that filmmakers may use this expertise to simply take away wires from images or exchange lacking pixels with the proper background.

On the software program facet, Nvidia additionally at present introduced that it’s open sourcing its Material Definition Language (MDL).

Corporations starting from Adobe (for Dimension CC) to Pixar, Siemens, Black Magic, Weta Digital, Epic Video games and Autodesk have already signed as much as assist the brand new Turing structure.

All of this energy comes at a value, after all. The brand new Quadro RTX line begins at $2,300 for a 16GB model, whereas stepping as much as 24GB will set you again $6,300. Double that reminiscence to 48GB and Nvidia expects that you just’ll pay about $10,000 for this high-end card.