Designing, simulating and citing trendy information facilities is extremely advanced, involving a number of issues like efficiency, power effectivity and scalability.
It additionally requires bringing collectively a workforce of extremely expert engineers throughout compute and community design, computer-aided design (CAD) modeling, and mechanical, electrical and thermal design.
NVIDIA builds the world’s most superior AI supercomputers and at GTC unveiled its newest — a big cluster based mostly on the NVIDIA GB200 NVL72 liquid-cooled system. It consists of two racks, every containing 18 NVIDIA Grace CPUs and 36 NVIDIA Blackwell GPUs, related by fourth-generation NVIDIA NVLink switches.
On the present flooring, NVIDIA demoed this totally operational information middle as a digital twin in NVIDIA Omniverse, a platform for connecting and constructing generative AI-enabled 3D pipelines, instruments, purposes and companies.
To deliver up new information facilities as quick as attainable, NVIDIA first constructed its digital twin with software program instruments related by Omniverse. Engineers unified and visualized a number of CAD datasets in full bodily accuracy and photorealism in Common Scene Description (OpenUSD) utilizing the Cadence Actuality digital twin platform, powered by NVIDIA Omniverse APIs.
Design, Simulate and Optimize With Enhanced Effectivity and Accuracy
The brand new GB200 cluster is changing an present cluster in one among NVIDIA’s legacy information facilities. To begin the digital build-out, expertise firm Kinetic Imaginative and prescient scanned the power utilizing the NavVis VLX wearable lidar scanner to supply extremely correct level cloud information and panorama images.
Then, Prevu3D software program was used to take away the present clusters and convert the purpose cloud to a 3D mesh. This offered a bodily correct 3D mannequin of the power, during which the brand new digital information middle may very well be simulated.
Engineers mixed and visualized a number of CAD datasets with enhanced precision and realism by utilizing the Cadence Actuality platform. The platform’s integration with Omniverse offered a robust computing platform that enabled groups to develop OpenUSD-based 3D instruments, workflows and purposes.
Omniverse Cloud APIs additionally added interoperability with extra instruments, together with PATCH MANAGER and NVIDIA Air. With PATCH MANAGER, the workforce designed the bodily structure of their cluster and networking infrastructure, making certain that cabling lengths had been correct and routing was correctly configured.
The workforce used Cadence’s Actuality Digital Twin solvers, accelerated by NVIDIA Modulus APIs and NVIDIA Grace Hopper, to simulate airflows, in addition to the efficiency of the brand new liquid-cooling programs from companions like Vertiv and Schneider Electrical. The built-in cooling programs within the GB200 trays had been simulated and optimized utilizing options from Ansys, which introduced simulation information into the digital twin.
The demo confirmed how digital twins can enable customers to totally check, optimize and validate information middle designs earlier than ever producing a bodily system. By visualizing the efficiency of the info middle within the digital twin, groups can higher optimize their designs and plan for what-if situations.
Customers also can improve information middle and cluster designs by balancing disparate units of boundary circumstances, akin to cabling lengths, energy, cooling and area, in an built-in method — enabling engineers and design groups to deliver clusters on-line a lot sooner and with extra effectivity and optimization than earlier than.
Study extra about NVIDIA Omniverse and NVIDIA Blackwell.
