The Next Leap in AI and CFD Simulation: Co-Optimization from Cadence and NVIDIA

Innovation in the AI and supercomputing domains is proceeding at a rapid pace, with each new advancement heralding a future more tightly interwoven with the threads of intelligence and computation. Cadence, with the release of its Millennium Platform, co-optimized with NVIDIA, underscores the paradigm shift underway, positioning this new breed of AI supercomputer at the leading edge of a multiphysics revolution.

The Cadence Millennium Platform provides near-linear scaling across applications, from external aerodynamics and aero-acoustics to combustion and multiphysics. It enables massive LES simulations with the fastest time to results—such as the accurate simulation of a realistic aircraft in landing configuration, in under 14 hours on just two GPU nodes instead of several days on traditional CPU-based systems.

The Cadence Millennium Platform unlocks the ability for engineers to simulate systems previously undreamed of without sacrificing accuracy, expanding data-driven analysis and optimization from the component level to the entire system—for automobiles, airplanes, data centers, and the surrounding environment.

The Intersection of AI and Multiphysics Simulation

Multiphysics simulation involves creating a single predictive model that incorporates various forms of physics, such as fluid dynamics, thermal, structural, acoustics, optics, electromagnetism, etc. This type of modeling requires the integration of knowledge across many separate disciplines. However, the increasing complexity of the systems and analyses requires more advanced computational capabilities to achieve the necessary accuracy and capacity without exploding the simulation time. The co-optimization of the principled simulation software to take advantage of modern accelerated compute and AI achieves this goal. For physics-ML workloads, NVIDIA Modulus accelerates Cadence’s Reality Digital Twin solvers to simulate airflows. Engineers can use AI surrogates trained with NVIDIA Modulus for “what-if” analysis that simulates the air and temperature fields almost instantly.

Cadence’s collaboration with NVIDIA ushers in a new era where the power of AI is combined with the strength of physics-based modeling for even more robust and accurate results. This redefines the boundary of what is computationally feasible.

Addressing the Insatiable Compute Appetite of AI

The increasing demand for computational resources to fuel advancements in AI technology cannot be emphasized enough. As the complexity of AI models continues to grow and the size of datasets expands, the need for high processing power, memory bandwidth, and interconnect speeds is skyrocketing. Enterprises seek systems that meet these insatiable demands while ensuring power efficiency and scalability.

For years, computational fluid dynamics (CFD) simulations have helped reduce the time needed for design processes and provided insights that support or even eliminate the need for physical experiments. However, the need for accuracy, speed, and better computing resources has limited their broader application in the past. With the introduction of advanced AI technologies, overcoming these challenges is crucial because traditional CFD simulations fail to generate sufficient data to train AI models effectively.

The Millennium Platform emerges as a pivotal solution by providing precise predictions for intricate CFD challenges, including flow separation, transition, turbulent mixing, and aero-acoustic noise. It employs a judicious combination of large-eddy simulation (LES) for meticulous detail complemented by high-throughput Reynolds-averaged Navier-Stokes (RANS) simulations for unparalleled speed.

The Integration of HW/SW Acceleration

By harnessing hardware acceleration technologies powered by cutting-edge NVIDIA GPUs, the Millennium Platform can significantly reduce computing costs by up to 10X and decrease energy consumption by up to 20X compared to traditional HPC frameworks. This remarkable feat enables the completion of extensive simulations in a fraction of the previous time, thereby enhancing the ability to generate voluminous, high-quality simulation databases automatically.

AI models can be trained efficiently using these large databases on the same Millennium hardware where they were created. Furthermore, the Millennium hardware can run these AI models, replacing some future simulations. Thus, a single Millennium M1 box can significantly speed up the CFD simulation process, create large datasets for training AI models, and run these models to enhance productivity further.

The Millennium Platform leverages a unique hardware/software acceleration platform and NVIDIA's GPU and CPU computing prowess, along with proprietary software algorithms, to realize an up to 100X improvement in design impact. This unparalleled accuracy, speed, and scalability amalgamate digital twin technologies, AI, and HPC to deliver tailored, highly effective multiphysics simulation solutions across key sectors, including automotive, aerospace and defense, energy, turbomachinery, and data centers. Furthermore, integrating generative AI technology into this blend markedly accelerates the design and analysis exploration, facilitating optimal system solutions through a deeper understanding of design intricacies.

The integration of Cadence's Millennium Platform and NVIDIA's H100/A100 Tensor Core GPUs, alongside the NVIDIA GH200 Grace Hopper and recently announced GB200 Grace Blackwell Superchips, heralds a new era of potentiality, underscored by remarkable performance metrics.

This synergy yields abundant, reliable training data and further pushes the frontiers of computational resource design and deployment. Significantly, these co-optimized systems play a critical role in enabling the effective training of AI models, setting a solid foundation for their application in advanced simulations.

Co-Optimized with the NVIDIA Accelerated Computing Platform

Cadence’s Millennium Platform leverages specialized NVIDIA hardware for accelerated computing, markedly improving efficiency. As a result, a single Millennium M1 box can perform at a level comparable to 32,000 CPUs, providing near-emulation speeds. In the context of CFD, emulation can be up to 1,000X faster than traditional simulation, representing a groundbreaking advancement across all three fundamental layers of this technology. By producing large volumes of accurate data, the Millennium Platform creates a foundation for AI to enhance simulation capabilities, leading to more efficient and high-quality outputs without simulating every possible scenario. This collaboration drives the demand for advanced computing and leads to the creation of superior computational solutions.

Integrating AI with principled simulation, boosted by innovative computing technologies, represents the core of Cadence’s innovative approach. With the NVIDIA accelerated computing platform at its core plus Cadence computational software, the Millennium Platform brings together these three concepts—AI, principled simulation + optimization, and accelerated compute—in one system, working in concert to push the boundaries of what is possible, cementing the foundation of this unique contribution to the field.

Deep Collaboration on EDA, SDA, Digital Biology, Omniverse Digital Twins, and Gen AI

The underlying applied mathematics driving the computational heft of the Millennium Platform speaks volumes about the synchronicity between pure research and its practical applications. Cadence is also adopting NVIDIA Omniverse Cloud APIs into its Reality Digital Twin Platform for data center design and simulation—enabling data center engineers to bring in multi-CAD datasets and simulate air and liquid-cooling system performance in full context. Cadence Molecular Sciences (OpenEye) is also integrating its Orion cloud molecular design platform with NVIDIA BioNeMo microservices to accelerate large and small molecule therapeutic design.

A Century-Defining Technology

With its ramifications across multiple industries, the release of the Cadence Millennium Platform, co-optimized with NVIDIA, is a page from the future of multiphysics modeling, addressing head-on the fundamental design challenges we face at the societal level.

READ MORE