Dell Technologies used the ISC conference in Hamburg, Germany, to introduce the PowerEdge XE8812, a new server developed in deeper collaboration with NVIDIA. The system is the first in Dell's lineup to use the NVIDIA Vera Rubin NVL4 architecture, and it is built around a 100% direct liquid-cooling, fanless design. Dell said a single rack can hold up to 144 GPUs, with the platform aimed at converging AI and high-performance computing workloads. Global availability is scheduled for early 2027.
Vera Rubin NVL4 pushes compute density higher
Dell positioned the XE8812 as a major step up for demanding AI and scientific computing jobs. Compared with the previous generation, the company said core count rises from 144 to 176, while memory capacity per socket and per GPU increases by 50%. That matters for large language models and complex simulation work, where keeping more workloads resident in memory can reduce the delays tied to staging and swapping.
Thermal design is central to the system. The XE8812 supports more than 300kW of power and pairs its liquid-cooled hardware with the Dell PowerRack 9100, which is based on OCP open standards. Dell said the setup is designed to maximize compute density while limiting physical footprint. With factory-integrated and pre-validated deployment services, the company added, installation and production launch can be completed in as little as 6 hours.
Dell says more than 5,000 organizations already use Dell AI Factory
Dell also tied the launch to existing market adoption. According to the company, more than 5,000 enterprises and institutions have already deployed Dell AI Factory. One of the best-known examples is Doudna, the next flagship supercomputer at the U.S. Department of Energy's Lawrence Berkeley National Laboratory. That system is based on the XE8812 and Vera Rubin NVL4 architecture, and uses NVIDIA Quantum-X800 InfiniBand networking to support research ranging from molecular biology to astronomy.
The infrastructure has also been adopted outside the United States. The report cited France-based InstaDeep's Kyber supercomputer for AI model training and automated design, with compute performance of 0.5 exaFLOPs. It also pointed to the Wellcome Sanger Institute in the UK for large-scale genome decoding, and Monash University's MAVERIC supercomputer in Australia for cancer and climate research.
Executives frame the system around AI-HPC convergence
Arun Narayanan, senior vice president for compute and networking at Dell Technologies, said research institutions working on major scientific problems need infrastructure that matches those ambitions. Chris Marriott, vice president of enterprise platforms at NVIDIA, said the convergence of AI and HPC is reshaping expectations for hardware performance and efficiency, with the Dell-NVIDIA partnership focused on delivering an open ecosystem for those workloads.
According to Dell's announcement, the PowerEdge XE8812 is set for a global launch in early 2027.