The next round of innovation in scientific computing, AI supercomputing and digital twins

With the large-scale cloud service providers, Internet giants and national supercomputing centers investing in the development of AI supercomputing, the application scope of such systems has been greatly expanded, such as digital twins. Even SDA manufacturer Cadence has joined the ranks of building AI supercomputing systems to develop digital twin applications.

AI supercomputing and digital twins
In the era of universal computing workloads in the past, it was difficult for us to adopt larger scale computing clusters to build digital twins. With the emergence of AI technology, especially generative AI technology, adopting highly customized AI supercomputing systems has become one of the new ways to quickly achieve digital twins. As a result, digital twins have been popularized in scientific calculations such as climate research, drug development, and gene sequencing.

Nvidia has built the Earth-2 cloud platform, a digital twin system targeting the Earth's climate, based on its AI supercomputing. Nvidia is also creating a new generative AI model called CorrDiff for this climate research and development platform. This model is based on diffusion simulation technology, which can increase the resolution of current numerical simulations by 12.5 times. Compared with traditional methods, it improves speed and energy efficiency by 1000 times and 3000 times, respectively. This model can be effectively used to study the physical characteristics of local weather in a small area, such as the landing place of typhoon.

Of course, the visualization part of digital twins also relies on the support of NVIDIA Omniverse platform. By combining meteorological data with Omniverse, users can visually perceive the impact of actual weather and provide weather simulations at higher resolutions.

Supercomputing for simulation of multi physics field systems
In addition to climate research, digital twin technology can also be used in other physical studies, such as fluid dynamics simulation, but such simulation tasks often require massive computational performance and can only be completed with the help of supercomputers. In February of this year, Cadence built a supercomputing platform called Cadence Millennium M1 for digital twins based on NVIDIA's GPU. The system can be used to provide high-performance, high energy efficiency, and high fidelity computational fluid dynamics simulations.

NVIDIA's high-performance GPU, combined with Cadence's high fidelity CFD software stack, has been optimized for GPU acceleration and generative AI. The computing power of each GPU is equivalent to the computing performance of up to 1000 CPU cores. The calculation iteration time has been reduced from weeks to hours, and compared to CPUs of the same level, energy efficiency has been improved by nearly 20 times.

In the design of automobiles and eVTOL, high requirements are placed on the energy efficiency of systems, such as improving fuel efficiency, reducing drag and noise, all of which involve computational fluid dynamics. With the support of GPU CFD solver, the iteration speed of related simulation applications is significantly accelerated, which helps manufacturing enterprises create better appearance designs. The addition of digital twins also provides support for high fidelity modeling. This is also the reason why Cadence has reached a technical cooperation with the F1 team, and the calculation and simulation of fluid dynamics can further improve the overall performance of the racing car.

Cadence, who has been cultivating the chip simulation software and hardware market for many years, previously provided powerful simulation computing power support for chip design with the Palladium platform based on FPGA. The Cadence Millennium M1, released this time, provides hardware support for multi physics field simulation and represents Cadence's comprehensive transformation from EDA to SDA in terms of software and hardware product solutions.

With the changes in market demand, traditional software vendors like Cadence in the past are gradually expanding their hardware and cloud services businesses, which is the transformation brought about by AI and digital twin technology. Compared to the large-scale supercomputing required for traditional HPC applications, smaller scale AI supercomputing may take a step ahead in commercial implementation.