First purpose-built supercomputer emulator for AI design
Cette publication existe aussi en Français
EDA vendor Cadence Design Systems has expanded its AI supercomputer using Nvidia’s Blackwell technology.
The Millennium M2000 Supercomputer is based on the Nvidia HGX B200 systems, RTX Pro 6000 Blackwell Server Edition GPUs and CUDA-X libraries integrated with the Cadence solver software.
This follows the Millennium M1 Supercomputer launched in February 2024 for computational fluid dynamics (CFD) simulations in automotive applications. This used GPUs from Nvidia and AMD, in the cloud with minimum of eight GPUs, or on premises with a minimum of 32 GPUs.
The HGX B200 is a liquid-cooled eight GPU system that supports up to 96 GPUs in a single rack. The RTX Pro is used for rendering, with 96GB of GDDR7 memory and support for Multi-Instance GPU, or MIG. This allows each RTX PRO 6000 to be partitioned into as many as four fully isolated instances with 24GB each to run simultaneous AI and graphics workloads.
The RTX Pro 6000 is also the first universal GPU to enable secure AI with NVIDIA Confidential Computing, which protects AI models and sensitive data from unauthorized access with strong, hardware-based security. This provides a physically isolated trusted execution environment to secure the entire workload while data is in use.
This dedicated Blackwell-based system provides up to 80X higher performance versus CPU-based systems for electronic design automation (EDA), system design and analysis (SDA), and drug discovery applications. It also includes support for a wider range of solvers and digital twin technologies.
This is the industry’s first purpose-built emulator for AI design with all the multiphysics capabilities needed to analyze and optimize 3D-IC and advanced packaging designs, including power, thermal, stress/warpage and electromagnetics. This improves the quality of results in less time, giving engineering teams can achieve greater reliability and efficiency in their product development cycles. For example, traditional semiconductor chip-level power integrity simulations are limited to small windows of time. Developers can now run simulations in less than a day that previously would have taken hundreds of CPUs almost two weeks.
The tightly co-optimized hardware-software stack also runs with 20X lower power consumption.
Customers include chip designer Mediatek and server maker Supermicro as well as supersonic aircraft developer Boom Aerospace.
- Honda has used the previous generation for air taxi research
- World’s largest FPGA boosts emulation system
“The Millennium M2000 Supercomputer will drive the next leap in AI-accelerated engineering by leveraging our massively scalable solvers, dedicated Nvidia Blackwell-accelerated computing and AI to help designers continue to push the limits of what is possible,” said Anirudh Devgan, president and CEO of Cadence. “Purpose-built for the most advanced AI models of today and tomorrow, the Millennium M2000 Supercomputer delivers unprecedented designer productivity to propel the next generation of AI infrastructure, physical AI systems and drug discovery.”
The system can also accelerate the design and operation of these datacentre digital twins and the modelling process required for the racks, boards and equipment that power them. The system also enables high-accuracy and high-capacity virtual simulations of autonomous transportation, drones and robotics.
The Millennium M2000 Supercomputer is available both in the cloud and as an on-premises appliance.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
