Nvidia expands Agent Toolkit with PhysicsNeMo and CUDA-X libraries

Nvidia (NASDAQ: NVDA) announced an expansion of its Agent Toolkit for engineering, adding re-architected PhysicsNeMo libraries and updated CUDA-X libraries to support autonomous AI engineering workflows in chip design, verification, packaging and systems.

The new additions give developers access to AI physics skills for training and deploying simulation models, accelerated sparse solvers through a new cuISS library, direct sparse solvers via cuDSS, and quantum chemistry capabilities through cuEST. These tools are designed to be integrated into agentic engineering workflows running on Nvidia GPUs.

Nvidia also said its Nemotron 3 Ultra open model leads among open models in agentic register-transfer level coding on the comprehensive verilog design problems benchmark, using the ACE-RTL agent developed by Nvidia Research. The model can be post-trained on proprietary data and deployed on premises.

Several companies said they are using Nvidia's expanded toolkit. Cadence is using Nvidia Nemotron, accelerated computing and CUDA-X libraries with its AuraStack AI Super Agent and Millennium M2000 platform, claiming up to 20x faster multiphysics performance. Siemens is using Nvidia NeMo Gym and Nemotron models with its Fuse EDA AI Agent, reporting more than 10x faster library characterization and more than 10x reduction in token costs in its Solido Characterization Suite.

Samsung is applying Nvidia PhysicsNeMo for chip-scale thermal-stress analysis across domains containing up to 10 billion cells, and using cuLitho and CUDA-X libraries for up to 20x greater computational lithography performance. Keysight said it is using cuDSS to accelerate electromagnetic simulations by up to 10x. Samsung, Synopsys and TSMC said they are integrating cuEST to achieve up to a 50x speedup for quantum chemistry workloads.

Synopsys is incorporating the Agent Toolkit and Nemotron models into its AgentEngineer platform for chip and system design workflows. Silvaco said it completed a 3.2-billion-mesh-node photonic simulation in under four hours using 32 Nvidia GPUs connected via NVLink.

Timothy Costa, vice president and general manager of computational engineering at Nvidia, said, "With NVIDIA Agent Toolkit, developers can build agentic engineers that reason using physics, run complex simulations and generate high-fidelity data."

You May Also Be Interested In

• Active options: NVDA TSLA AAPL AMZN MSFT MU META GOOGL
• FAA grants Boeing 737-7 amended type certificate
• Leidos to supply infrared sensors for Sierra Space missile defense sats