NVIDIA Apollo — a family of open models for accelerating industrial and computational engineering — was introduced today at the SC25 conference in St. Louis.
Accelerated by NVIDIA AI infrastructure, the new AI physics models will enable developers to integrate real-time capabilities into their simulation software across a broad range of industries.
The NVIDIA Apollo family will include physics-optimized models — each developed for scalability, performance and accuracy — for fields including:
The family of open models harness the latest developments in AI physics, incorporating best-in-class machine learning architectures, such as neural operators, transformers and diffusion methods, with domain-specific knowledge.
NVIDIA Apollo will provide pretrained checkpoints and reference workflows for training, inference and benchmarking, allowing developers to integrate and customize the models for their specific needs.
Applied Materials, Cadence, LAM Research Corp., Luminary Cloud, KLA, PhysicsX, Rescale, Siemens and Synopsys are among the industry leaders that intend to train, fine-tune and deploy their AI technologies using the new open models. These companies are already using NVIDIA AI models and infrastructure to bolster their applications.
Applied Materials is developing new materials and manufacturing processes with NVIDIA AI physics to improve the power efficiency of both the manufacturing process and the final product, directly addressing the most significant limiter in scaling semiconductor manufacturing capacity.
With NVIDIA GPUs and the CUDA framework, Applied has achieved up to 35x acceleration in modules of its ACE+ multi-physics software, enabling faster exploration and optimization of the semiconductor processes. Using ACE+ physics data, Applied has built AI models for key material modification technologies, enabling near-real-time flow, plasma and thermal modeling of advanced semiconductor process chambers using surrogate models — AI models trained on data from conventional simulations that can predict new cases in just seconds — and digital twins.
Cadence used its Fidelity Charles Solver, which is part of its Fidelity CFD software and is accelerated using the NVIDIA-powered Millennium M2000 Supercomputer, to produce a high-quality dataset of thousands of detailed, time-dependent full aircraft simulations. This data was used to train an AI physics model that enabled a real-time digital twin of a full aircraft, which was showcased last month at NVIDIA GTC Washington, D.C.
LAM Research is working with NVIDIA to accelerate plasma reactor simulation using NVIDIA AI physics. Plasma reactors are critical for etching and deposition processes in semiconductor manufacturing.
KLA will explore using NVIDIA Apollo models to accelerate a range of simulations. Faster, more accurate simulations will build on KLA’s existing capabilities and accelerate its development of new semiconductor process control solutions.
Northrop Grumman and Luminary Cloud are also using NVIDIA AI physics models to accelerate spacecraft thruster nozzle design. Harnessing NVIDIA CUDA-X libraries to accelerate its CFD solver, Northrop Grumman generated a large training dataset to build a surrogate model for nozzle simulation on Luminary Cloud’s platform, which is powered by NVIDIA AI physics models. This AI physics model will enable Northrop Grumman’s engineers to rapidly explore thousands of designs in record time.
PhysicsX’s AI-native platform supports the complete AI lifecycle, from simulation and data management to model training, fine-tuning and deployment, seamlessly integrating with NVIDIA AI physics infrastructure and simulation software like Siemens Simcenter X. For customers in automotive, aerospace, energy and more, the PhysicsX platform dramatically reduces product development cycles and accelerates time to market.
Rescale is accelerating engineering innovation by integrating NVIDIA Apollo models into its industry-leading AI physics operating system. This enhancement to Rescale’s complete, end-to-end stack will allow engineers to seamlessly blend high-fidelity, first-principles simulations with high-speed AI surrogates. By using the advanced capabilities of NVIDIA Apollo models within the Rescale framework, customers will be able to explore vast design spaces orders of magnitude faster and achieve real-time inference results while maintaining the accuracy of traditional simulation methods.
Siemens is integrating NVIDIA AI physics into its flagship fluid simulation tools like Simcenter STAR-CCM+. This integration allows designers to blend high-fidelity first principles simulations with high-speed AI surrogates. This allows exploration of design options orders of magnitude faster than previously possible.
Synopsys is using NVIDIA AI physics to multiply GPU acceleration and achieve up to 500x speedups in computational engineering. The runtime of NVIDIA GPU-accelerated fluid simulation tools like Ansys Fluent can be greatly reduced by initializing the simulation with AI physics surrogates. This approach is faster than initializing simulations with traditional methods.
NVIDIA Apollo models are coming soon to build.nvidia.com, HuggingFace and as NVIDIA NIM microservices. Sign up to be notified when they’re available.
See notice regarding software product information.
