Nvidia boasts 50x faster simulation at GTC 2025

The chipmaker says its Blackwell processors have led to “an inflection point in engineering design.”

By Michael Alba

Welcome to Engineering Paper. Nvidia’s annual GTC conference is taking place this week in San Jose, California, and with it came the usual torrent of Nvidia news.

I can’t cover it all here, but you can check out Nvidia CEO Jensen Huang’s opening keynote for two hours of the chipmaker’s strategic vision, a heap of product announcements, some special stage props, and a few self-aggrandizing video interludes.

On the simulation side of things, one of Nvidia’s top announcements was really more of a brag: the company says its Blackwell chips (which were announced at last year’s GTC) are accelerating simulation software by up to 50x.

“We saw up to 50x better performance on a Blackwell chip as compared to a leading data center CPU,” Timothy Costa , senior director of CAE and CUDA-X at Nvidia, told me. “This is across a variety of important CAE workloads, from CFD to discrete element methods, finite element analysis, lithography and SPICE simulation.”

Nvidia’s announcement calls out a who’s who of CAE developers that have accelerated their software with Blackwell: Altair, Ansys, BeyondMath, Cadence, Comsol, Engys, Flexcompute, Hexagon, Luminary Cloud, M-Star, Navasto, Neural Concept, nTop, Rescale, Siemens, Simscale, Synopsys and Volcano Platforms, to put it alphabetically.

One concrete example comes from Cadence, which used a Blackwell-based server to run a 10 billion cell aerodynamic simulation of an aircraft during takeoff and landing.

“This is a problem that previously required a TOP500 supercomputer with hundreds of thousands of CPU cores running for days,” Costa said. “But this run was done on a single [Nvidia GB200] NVL72 server in under 24 hours.”

Nvidia concurrently announced the that its Omniverse Blueprint for real time digital twins, first previewed last year, is now generally available. (It’s also called OV RTDT, which my brain can’t help but read as R2D2.)

“The word Blueprint really means open source demo,” Costa said, and this one is meant to help Nvidia’s partners implement real-time digital twins. If you’re at GTC you can see some examples for yourself.

“At the show this week you'll see real time digital twins of cars, of supersonic jets, of the human heart—that’s a really cool one—and then many other incredible applications from our partners and their customers,” Costa said.

(I’m not at the show this year, so if you see any of those things send me your pictures, thoughts and maybe a fun postcard: [email protected].)

So what’s the bottom line of all this boasting?

“Nvidia superchip architectures, combined with advances in AI physics, have created an inflection point in engineering design,” Costa said. “Grand challenges that [were] previously too complex and costly are being incorporated into typical design cycles, and interactive design with real-time digital twins are becoming a reality.”

Speaking of AI physics…

Geometry to simulation to AI

AI-based design optimization is the focus of a new integration between Luminary Cloud, nTop and Nvidia.

Luminary Cloud announced that its APIs can now create a pipeline between its GPU-based simulation platform, nTop’s computational design capabilities and Nvidia’s PhysicsNeMo framework for physics-based AI.

Together the three tools allow users to automatically create geometry, analyze it, and use that data to train predictive AI models.

“NTop generates the geometry and all the parametric changes. You could generate thousands of geometries. Those geometries are fed directly into Luminary [Cloud], which analyzes the physics and produces results,” Juan Alonso , CTO and cofounder of Luminary Cloud, told me.

“Then, leveraging the Nvidia NeMo ecosystem to train models… you could use that model in lieu of the full simulation. Even though we’re very fast, inference from these models is even faster.”

While both Luminary Cloud and nTop offer generic geometry tools, Alonso said the integration will particularly benefit users that routinely rely on fluid or thermal analysis, such as in the automotive and aerospace industries. The developers demonstrated the integration by optimizing the lift and drag characteristics of a flying wing (see below image).

Bradley Rothenberg , CEO of nTop, provided a few more details and images in a LinkedIn post yesterday.

This is the first official collaboration between Luminary Cloud and nTop, but both companies have worked with Nvidia before. Luminary Cloud and Nvidia jointly demonstrated a virtual wind tunnel last November when Nvidia announced Omniverse Blueprints for real-time digital twins (now generally available; see above item). Last September, nTop announced a separate Nvidia integration and an investment from Nvidia’s venture capital arm, NVentures.

Ansys integrates Omniverse

In other Nvidia news, Ansys announced that it will integrate Nvidia Omniverse in some of its simulation software, starting with Ansys Fluent for fluid simulation and Ansys AVxcelerate Sensors for sensor simulation. Other Ansys apps will follow, according to the developer.

The Omniverse integration will allow Ansys users to render photorealistic models directly in the Fluent or AVxcelerate Sensors interfaces, which Ansys says will facilitate simulation data preparation and communication. PyAnsys, Ansys’ collection of Python packages, will further allow users to automatically format simulation data for their own applications built on Nvidia Omniverse.

“The integration of Omniverse technologies within Fluent allows us to visualize complex physics simulations that give us and our customers intuitive insight into how our equipment operates in stunning detail,” said Andrew Hobbs, director of advanced technologies at Astec Industries, in Ansys’ press release.

One last link

Are ants smarter than humans? The evidence is mounting. Read Mark Jones’ Technical thinking: Finally, proof that the Andy Letter was right to prepare for the upcoming war with Paratrechina longicornis.

Got news, tips, comments, or complaints? Send them my way: [email protected].

For more design and simulation stories, visit Engineering.com.