Member-only story
Digital Twins Will Take All The Computing Power They Can Get
The digital twin represents one corner of the artificial intelligence (AI) world that is going to account for a disproportionate use of future computing resources. It’s already pulling hard on existing ones.
A digital twin is pretty much what it sounds like, with a few special nuances. Initially used to design new products without a heavy upfront investment in physical prototypes, the digital twin started as a model of the physical product in digital space. Early prototypes included things like jet engines and aircraft bodies, applications in which physical models were extremely expensive and design failures could be catastrophic. It makes sense: build the digital version with all the design’s physical properties codified, and then play that model against data that represents the forces it will encounter (e.g., pressure, gravity, torque, wind, snarge, water). If the digital airplane breaks to pieces, no one loses a year’s worth of work, a steep investment, or their life.
Which brings up the nuances. The formal description of a digital twin specifies three subtypes:
- a prototype,
- an instance, and
- an aggregate.
In simple terms, the prototype is the fully developed model, with all its complexities, deployed during the design process. This is where the digital twin draws maximally on available computing resources. Typically, a digital twin prototype is run on a high performance computing (HPC) system. Such systems need a tremendous amount of single-threaded performance of the type touted by Nvidia CEO Jensen Huang during the company’s Global Technology Conference earlier this spring, when he praised Intel’s Ice Lakec Xeon server processors during his lecture on Omniverse. Describing Omniverse as “digital twins, virtual worlds, and the next evolution of the Internet,” Huang told of how these powerful models would live at the “edge;” that is, still in the cloud, but close to either the source or use of the data. Normally, Huang would not use the bully pulpit of his own conference to laud the quality of one of his competitor’s products, but, for the moment, single-threaded performance is what is needed to produce digital twin prototypes, and Nvidia’s strong suit is in parallel performance. More on that…
