Nvidia Turing

2018: Premiere of Nvidia Turing

On August 13, 2018 the NVIDIA company provided graphic architecture of NVIDIA Turing.

According to the statement of NVIDIA, the architecture of Turing received RT cores for acceleration of ray tracing and tensor cores for an inferens. Two engines – in combination with more productive calculations for simulation and the improved rasterization – give rise to generation of hybrid rendering for the industry of visual effects. Hybrid rendering allows to create content of cinema quality and effects based on neuronets and to provide the interactivity level unavailable earlier during the work with models of high complexity.

The NVIDIA company provided graphic architecture of Nvidia Turing

The company also provided the products based on architecture of Turing – graphic processors NVIDIA Quadro RTX 8000, Quadro RTX 6000 and Quadro RTX 5000.

In order that developers could involve completely this potential, NVIDIA added to the platform of development RTX SDK with tools for AI, ray tracing and simulation. The company also announced that key graphic applications which many designers, artists and scientists use will get access to functionality of Turing through the RTX platform.

The ray tracing in real time accelerated by RT cores

The architecture of Turing is equipped with special processors for ray tracing which are called RT cores. They accelerate calculations of the movement of light and a sound in 3D - Wednesday to 10 GigaRays per second. Turing accelerates ray tracing to 25 times in comparison with the previous generation of Pascal, and GPU ноды cope with final rendering when imposing effects in movies 30 times faster, than notes based on GPU, the developer noted.

AI with acceleration by tensor cores

According to information provided by the developer, the architecture of Turing also locates tensor cores – processors which accelerate training of deep networks and inferens, providing up to 500 trillion tensor operations per second. Such performance level gives possibilities of AI for creation of powerful applications. DLAA — smoothing based on methods of deep learning treats them. This method works in generation of qualitative dynamic images and also in noise elimination, scaling of permission and video a retayming.

These opportunities are included into NVIDIA NGX - a packet of tools of the developer based on algorithms of deep learning which allows developers to integrate graphics, imposing of photos and processing of video into applications with the trained networks, noted in NVIDIA.

The accelerated simulation and rasterization with the strimingovy Turing multiprocessor

GPU based on Turing are equipped with the strimingovy multiprocessor (SM) which adds the integer executive block in parallel to the channel of data with a floating point, and the unified architecture of a cache with the bandwidth doubled in comparison with the previous generation.

In combination with such graphic technologies as Variable Rate Shading, the strimingovy Turing multiprocessor reaches high performance on a core. Locating 4608 cores of CUDA, Turing supports up to 16 trillion transactions with a floating point from 16 trillion integer operations per second, claim in NVIDIA.

Developers can use NVIDIA CUDA 10, FleX and PhysX for creation of difficult simulations, such as particles or dynamics of liquids, for scientific visualization, virtual environments and special effects.

GPU Quadro based on Turing will appear in the market in the fourth quarter 2018.