At Siggraph 2018, Nvidia has announced a new Turing architecture and Quadro RTX cards based on it that the company says will revolutionise how 3D artists work.
Turing is the successor to Pascal, which is used by Nvidia's current generation of graphics cards that top out with the Quadro P6000. Nvidia says that Turing is up to 6x faster at simulating the physical world – and will allows real-time raytracing within industry-standard 3D applications and accelerated output rendering.
The company isn’t holding back at how much of an innovation it thinks the new architecture is, with CEO Jensen Huang describing it on stage at a launch keynote at the Siggraph animation conference as “the greatest leap since the invention of the CUDA GPU in 2006”.
This is because the platform allows real-time raytracing, which was previously only possible using the company’s Volta platform, which is for datacentres, not desktop PCs. Now, claims Nvidia, artists will be able to see a much more accurate preview – perhaps completely accurate – of their final artwork or animation.
Photo-realistic raytracing in real-time with global illumination has been the goal of Nvidia for 12 years, said Jensen, but previously the company predicted that it would five to 10 years away from now. It can be done now because Nvidia has developed specific ray-tracing cores that are used in a new line of Quadro graphics cards called the RTX line, along with Tensor cores for machine-learning-based operations such as noise reduction.
This is the secret to the claim of real-time ray-tracing. The output from the ray-tracing cores is noisy, but this is then run through the machine-learning based noise-reduction algorithm to produce a very high quality render.
The first Quadro cards based on the Turing architecture will be the RTX 5000, RTX 6000 and RTX 8000.
The RTX 8000 is the flagship card, with 4608 CUDA cores and 576 Tensor cores. It has 48GB of GDDR6 memory, which can be doubled to 96GB by using two cards in the same workstation through the new NVLink 2.0 interface. GDDR6 has twice the bandwidth of the GDDR5 memory used by the Pascal-based cards.
For ray-tracing, the RTX 8000 can output 10 Giga rays per second, which sounds impressive but we really don’t have anything to compare it with.
The RTX 6000 is essentially the same card, but with 24GB of RAM (48GB through NVLink 2.0). The RTX 5000 has 3,072 CUDA cores, 384 Tensor cores and 16GB of RAM – with a ray-tracing capability of 6 Giga rays/s.
The new Quadro RTX cards will ship in the fourth quarter of 2018, costing around US$2,300 (around £1,800) for the RTX 5000, $6,300 (£4,940) for the RTX 6000 and $10,000 (£7,830) for the RTX 8000.
All support the new mooted VirtualLink standard for connecting VR headsets over a USB-C lead.
Alongside the cards, Nvidia has announced RTX Server, which combines eight RTX cards with the company’s Infinity software to created render farm datacentres and virtual workstations. Some customers will get early access in Q4 of this year, with it being generally available in Q1 2019.
Nvidia claims that the RTX Server will cost about a quarter of a CPU-based render farm, taking up around a tenth of the space and using around an eleventh of the power. Or conversely, for the same money you can get four times the performance.
Application support for Nvidia Quadro RTX
To take advantage of the power of these cards and the Turing architecture, application developers need to add support for the RTX platform. At the time of launch, Nvidia says it has support from apps including Adobe Dimension, Autodesk Arnold, Allegorithmic Substance Designer and Project Alchemist, Blackmagic Design Resolve 15, Epic Games Unreal Engine, OTOY Octane Render and Pixar Renderman – with most of these in beta and being demoed on the show floor at Siggraph.
One of the first to show off real-time rendering on RTX boards is Chaos Group with its new Project Lavina renderer. You can see it in action in the video below.
We'll be spending time with many of these companies over the next few days at Siggraph and will update this story with their thoughts on the RTX platform.