Dassault Systèmes and Vertiv will validate designs and performance in a virtual environment before construction to bring artificial intelligence data centers into operation sooner. Their goal is to reduce problems during on-site deployment by checking in advance whether power and cooling systems work together, cutting infrastructure deployment time by up to half.
According to Dassault Systèmes on the 22nd, local time, the companies announced a collaboration to support the design and construction of Nvidia DSX AI factories. Dassault Systèmes will support design and simulation through its 3DEXPERIENCE platform, while Vertiv will provide power and cooling infrastructure technology.
An AI factory is a data center equipped with computing infrastructure for AI training and inference. The companies explained that power, cooling, networking and computing equipment are closely interconnected, so changing one component can affect the performance of other systems. Their approach goes beyond checking individual pieces of equipment to verifying that the entire system works together.
The companies will use virtual twin technology to represent physical facilities in a virtual environment. Engineers can change equipment configurations to meet customer requirements and verify whether they can supply the necessary power and remove heat. They can also assess the effects of design changes on other equipment within the same environment.
Dassault Systèmes will incorporate information about Vertiv’s pre-engineered power and cooling equipment into its platform to support comparisons of different configurations. The companies will extend the model-based design approach used for Vertiv’s SmartRun to OneCore, a data center solution integrating power and thermal management equipment.
They will use Nvidia’s Vera Rubin DSX AI Factory reference design and Omniverse DSX Blueprint as design standards. They will support consistent management of design information from initial design through construction, commissioning and future equipment replacement.
The companies set goals of increasing the speed of configuring and validating infrastructure to meet customer requirements by up to 10 times and reducing the scheduled infrastructure deployment period by up to 50%.
Vertiv estimated through its own modeling that facilities could begin operating approximately 10 weeks earlier than with conventional methods. Its calculations suggest that, even with the same capacity, earlier operation could increase cumulative AI token output by up to 30% during the first 12 months after project initiation. Tokens are units used to count data processed or generated by AI.
Related articles
Vertiv said the figures were modeling results based on specific assumptions, not measured outcomes from actual customer environments. It explained that they reflect only the effects of bringing facilities into operation earlier and do not indicate improved operating efficiency.
“Competitive advantage in the AI era depends on how quickly computing capacity can be converted into actual production capability,” said Scott Armul, Vertiv’s chief product and technology officer. “This requires the entire AI factory, encompassing computing, power, cooling and physical infrastructure, to operate as a single system.”
Original Korean article: ZDNet Korea
English brief by Korea Wind Intelligence, koreawindintel.com
