NVIDIA introduces new physics module for Omniverse!


NVIDIA continues to develop its platform called Omniverse, which it introduced a while ago. Here are the details….


Artificial intelligence solutions continue to make our lives easier. Today, many jobs can be done through digital assistants and more advanced systems. NVIDIA, which is one of the companies that works the hardest in this field, offers both individual and corporate solutions thanks to the technologies it has developed.

Finally, the company, which presented these solutions at its event held today, made a show of strength. Giving an example from a project made by several companies, the US technology giant showed how important artificial intelligence-supported Omniverse is in power generation facilities or in planning the 5G infrastructure of an entire city.

NVIDIA shapes the future with Omniverse

Omniverse, which entered our lives last year, is the company’s real-time graphics simulation platform. Thanks to the project, simulations similar to real life and physically quite accurate can be created in the digital environment. Here is the work done with NVIDIA Omniverse.

Scientific studies accelerate with NVIDIA Physic-ML module

Working on solutions to be used in various fields, NVIDIA introduced a new physics-based toolkit to help engineers and scientists. The company, which supports molecular studies to accelerate drug studies, will also eliminate global challenges such as climate change with this project.

The module trains neural networks to use the fundamental laws of physics to model the behavior of complex systems in a wide variety of fields. It then uses it in a variety of digital twin applications, from industrial use cases to climate science.

Like most AI-based approaches, Physic-ML includes a data preparation module that helps manage observed or simulated data. It also describes the geometry of the systems it models and the explicit parameters of the space represented by the input geometry. Possible uses are as follows:

  • Sampling planner that allows the user to choose an approach such as semi-random sampling or substantial sampling to improve the convergence and accuracy of the trained model.
  • Python-based APIs for retrieving symbolic management partial differential equations and building physics-based neural networks.
  • Curated layers and network architectures that have proven effective for physics-based problems.
  • Physics-ML engine that takes these inputs to train the model using PyTorch and TensorFlow, cuDNN and multi-GPU for GPU acceleration, and NVIDIA Magnum IO for multi-node scaling.
  • The GPU-accelerated toolkit provides faster insights with fast turnaround that complements traditional analysis. The module allows users to explore different configurations and scenarios of a system by evaluating the impact of changing its parameters.

The high-performance TensorFlow-based application with modules optimizes performance by leveraging XLA, a domain-specific compiler for linear algebra that speeds up TensorFlow models. It uses the Horovod distributed deep learning training framework for multi-GPU scaling.

The last project to use it was a next-generation energy reactor. The researchers, who digitally modeled the reactor exactly, took advantage of this to identify potential problems. Thus, 1.6 billion dollars were saved from the money spent on annual maintenance.

Ericsson built city using Omniverse

Everything from the location of trees to the height and composition of buildings is crucial, as they affect 5G wireless signals in networks serving smartphones, tablets, and millions of other internet-connected devices.

The Stockholm-based company has combined decades of infrastructure and networking expertise with NVIDIA Omniverse Enterprise. By constructing a digital scale of a city they have contracted for 5G infrastructure, Ericsson created a realistic city simulation where everything from cars to trees to even the type of material used in buildings is calculated.

This way, the company won’t have to experiment with anticipating and solving environmental and motion-related problems. While this will increase efficiency, it will cause serious savings.