Ansys Collaborates with Supermicro and Nvidia for AI Hardware Solution

Ansys, in partnership with Supermicro and Nvidia, has developed custom hardware solutions aimed at significantly enhancing simulation speeds. This collaboration has resulted in hardware optimized specifically for Ansys multiphysics simulations, offering the potential to increase speeds by up to 1,600 times.

The Nvidia Supermicro hardware is designed to cater to Ansys multiphysics simulations and provides users with flexible configurations. These simulations are applicable in various industries, including automotive crash testing, external aerodynamics, aerospace gas turbine engines, and 5G/6G antennas.

Multiphysics simulation involves the integration of different physics solvers with various hardware options, each with its own performance advantages. Selecting and configuring the appropriate hardware for multiphysics simulation is a complex process that can significantly impact performance, cost, and productivity.

The customized Supermicro hardware, equipped with interconnects and cooling modules, has been utilized in testing scenarios. This hardware includes H100 GPUs, L40S GPUs, and the Nvidia Grace CPU Superchip, enabling engineers to conduct predictively accurate simulations more efficiently.

In a significant development, Synopsys has expressed interest in acquiring Ansys in a deal valued at $35 billion. This potential acquisition could have far-reaching implications for the simulation software industry and the technologies developed by both companies.

Furthermore, Ansys has collaborated with BMW to co-develop simulation software tailored for automated and autonomous driving applications. This partnership underscores the importance of simulation tools in advancing the capabilities of autonomous vehicles and enhancing safety standards in the automotive sector.

During collaborative testing, it was discovered that achieving the performance of Ansys Fluent and Rocky tools running on a single Nvidia GPU would require a substantial number of CPU cores. For instance, the Perceive EM tool running on one Nvidia GPU demonstrated performance equivalent to over a million CPU cores.

The testing process revealed remarkable speed improvements, with optiSLang AI+ achieving a 1,600x speed-up, Fluent showing a 24x improvement, Rocky a 17x enhancement, and the LS-Dyna tool a 4x boost. These advancements highlight the efficiency gains made possible by the collaboration between Ansys, Supermicro, and Nvidia.

Shane Emswiler, senior vice president of products at Ansys, emphasized the benefits of Supermicro's architecture for Ansys solutions, stating that it reduces constraints on simulation models' size and complexity. He also highlighted Ansys' commitment to transitioning solutions from Hopper chips to Blackwell, offering advanced configurations for a wide range of physics applications.

Dion Harris, director of data center product solutions at Nvidia, emphasized the role of simulation solutions in developing cutting-edge AI superchips. He noted that Nvidia's data center AI and Digital Twin platforms are pushing the boundaries of simulation performance, with Ansys already leveraging Nvidia's Omniverse tool for 3D multiphysics visualizations.

For more information, visit the official websites of Ansys (www.ansys.com), Nvidia (www.nvidia.com), and Supermicro (www.supermicro.com).