NVIDIA Announcing “the Most Powerful GPU RTX” on SIGGRAPH 2018

Vancouver, Canada – From 4 to 6 pm on August 13^th, on SIGGRAPH 2018, the founder and CEO of NVIDIA , Jensen Huang, announced the first ray-tracing GPU in the world. He claimed this GPU the most important one ever since 2006, which took 10 years to make.

SIGGRAPH 2018 is the 45^th Conference of SIGGRAPH, held from August 12^th to  16^th in Vancouver. SIGGRAPH is the most influential and the largest international conference and academic symposium on CG and interactive techniques organized by ACM SIGGRAPH, representing the highest level of Graphics and combining science, commerce and art.
As a professional spokesman of the Leather Jacket from Silicon Valley, Huang Renxun once again came on stage in his leather jacket. After retrospecting the development process of CG techniques of real-time ray tracing, CPU and GPU, he announced the publication of the latest ray-tracing GPU – Quadro from RTX Family.

The Most Powerful GPU in History, Renovating Image Rendering
Today, Jensen Huang announced the publication of 1^st ray-tracing GPU in the world – Quadro RTX GPU, which he claimed the most important GPU after the announcement of CUDA in 2006 and renovated the Computing of image-rendering. To invent this GPU, NVIDIA has been spending 10 years on relevant research.

On GTC 2017 held by NVIDIA this May, NVIDIA needed a DGX-Station super computer that carried 4 Tesla V100 to achieve the computing ability required by ray-tracing, yet it takes merely one GPU to complete now.
According to Jensen, the exterior of the GPU was designed to be smooth, thus reflecting the glittering lights on the spot of the conference, which accords with “Ray-tracing” in its name.
This GPU has 3 types of RTX – RTX 5000/6000/8000, 16 TFLOPS+16 Tips, 10 Giga Rays/sec, 500 Trillion Tensor Ops/sec and 100 GB/sec with NVLink.

Its core area rose rapidly to 754 square millimeters, with the largest VRAM 48GB GDDR6, the integration 18.6 billion transistors, the frequency of its VRAM 14GHz, enabling AI, ray-tracing and simulation of the acceleration of the hardwares.
There is one accident in the release conference when Jensen wanted to discuss a detail in his PPT with the staff member and the whole audience heard Jensen asking the staff, “Pardon? I couldn’t catch you…Well, I still couldn’t catch you.”
To break the embarrassment, Jensen joked to the audience, “Look, it shows that our speeches have not been previously practised. All our scenes are “real time” without being rendered in advance.”

Jensen said, now we could render the final CG effects of films with the newest GPU, which was unprecedented in history. On the spot, he displayed severalframes of the scene of the Spider-Man leaping and spinning in Avengers: Infinity War rendered by Quadro RTXs.
What is Ray-tracing?

RTX ray-tracing is a new technology announced by NVIDIA on GDC in March. It can calculate the routes of reflection, scattering and refraction of rays accurately and render photographic pictures that could hardly be distinguished from the real world.

In real world, we see 3D objects being illuminated by a light source, with the light particles bouncing from one object to another before reaching the eyes of the observer. By contrast, the technique of ray-tracing traces the light from our eyes (like a view camera) to capture these effects, tracing the paths of light of every elementary area on the surface of 2D images and applying them to 3D models of the scenes.
It is imaginable that the calculated amount of the technology is huge. Normally it takes several days or even weeks to render complicated special effects, so this technique is merely used to produce large-budget films.
However, with the emergence of “the Most Powerful GPU in history”, the calculation of rays is becoming increasingly simple. Within just a few minutes or even seconds, product designers, game designers and architects could produce photographic product models.
By the end of the speech, Jensen listed the leading factories that design creative applications and support RTX, including the image giants like Adobe, unity and VMD.
A New Generation of Framework – “Turing”

The core framework of ray-tracing GPU is the new framework announced by NVIDIA this time – Turing.

Previously on June 23th, NVIDIA celebrated the birthday of Alan Turing – the Father of Computer – on Tweet, which led to the masses speculating that NV did this deliberately to imply the potential name “Turing” of the framework of the next generation of its GPU. Now it has become the truth.
The framework “Turing” provides a computing performance of 16 TFLOPS+16TIPS and owns 4608 CUDA cores that could achieve fast physical simulations of the complicated real world. It will multiply the speed of its ray tracing 25 times and reach 30 times the speed of CPU.

On the spot, Jensen Huang presented the comparison pictures of the RTX with and without the framework “Turing”.
After presenting the Porsche demo based on RTX, Jensen announced the publication of Quadro RTX Server produced by NVIDIA. The server sets new standards for the rendering of the data center and easily arranges the panel points of the rendering, which enables massive and interactive rendering.
It combines Quadro RTX GPU with the new software Quadro Infinity which is to be announced in the first quarter next year, providing a powerful and flexible framework that meets the requirements of creative professionals. Quadro Infinity allows multiple users to visit a single GPU through the virtual work station, thus improving the density of the data center. Eventually users can easily arrange the rendering pitch points and work stations according to their specific requirements.
Quadro RTX server previously installs the leading software for content production and rendering in the industry, providing a powerful rendering solution that is easy to deploy. It could expand from mini-installation to the largest data center, only taking up one quarter of the cost of CPU in the rendering farm.

In the end, Jensen Huang introduced the prices of several types of Quadro RTX. Quadro RTX 8000 matches a memory of 48GB and costs 10000 dollars; Quadro RTX 6000 matches a memory of 24GB and costs 6300 dollars; Quadro RTX 5000 matches a memory of 16GB and costs 2300 dollars.
Quadro RTX GPU will be announced in the official website of NVIDIA in the 4^th quarter. Developers can reach these Quadro RTX through new NVIDIA RTXs (TM). RTX provides a series of creative applied programs and tools, including API that is used in ray-tracing, AI, rasterization, simulation, and easily accessible to professionals across industries.