Dell PowerEdge XE8812: Inside the New Vera Rubin AI Server and the DOE’s Doudna Supercomputer

• Architecture: Nvidia Vera Rubin NVL4, combining Nvidia's next-generation Arm-based Vera CPUs with Rubin GPUs .
• GPU density: Scales up to 144 GPUs per rack within the Dell PowerRack 9100, an OCP-standards-based rack architecture .
• Cooling: Fanless, direct liquid-cooled design — no large fan assemblies are needed .
• Performance per NVL72 rack configuration:
  • Up to 3.6 exaFLOPS (NVFP4) AI inference performance .
  • 2,520 PFLOPS (NVFP4) training performance .
  • 20.7 TB of HBM4 memory with 1,580 TB/s memory bandwidth .
  • 260 TB/s NVLink 6 interconnect bandwidth .
• Interconnect: NVLink 6 for ultra-high-speed GPU-to-GPU communication .

A related but higher-density variant, the PowerEdge XE9812, was already shipped as an NVL72 rack-scale system packing 72 Rubin GPUs and 36 Vera CPUs in a single liquid-cooled rack .

Industry Rollout Plans

The rollout of Vera Rubin-based servers began even before the official ISC announcement. On May 31, 2026, Dell shipped the world's first operational, liquid-cooled Vera Rubin NVL72 rack (the PowerEdge XE9812) to AI cloud provider CoreWeave . This marked the transition from announcement to real-world deployment.

The PowerEdge XE8812 and related Vera Rubin servers are part of a broader production rollout scheduled for the second half of 2026, following Nvidia's announcement at GTC 2026 that the Vera Rubin platform is in full production . Super Micro also launched competing Vera Rubin-based servers on the same day as Dell's ISC announcement .

Role in the Next DOE Supercomputer: Doudna

The Vera Rubin architecture directly powers the U.S. Department of Energy's next flagship supercomputer, named Doudna (also referred to as NERSC-10), to be housed at Lawrence Berkeley National Laboratory . The contract was awarded to Dell Technologies by the DOE in May 2025 .

Key details about the Doudna system:

• Performance target: Approximately 10× performance improvement over the current NERSC system (Perlmutter), with 3–5× better power efficiency . Some early estimates suggested 3–5× faster performance while consuming only 2–3 times more power .
• Deployment timeline: Expected to come online in 2026 .
• Scientific use cases: Fusion energy research, astronomy, life sciences, and CRISPR-related gene-editing discovery — the system is named after Nobel laureate Jennifer Doudna . More than 11,000 researchers are expected to use the system .
• Clarification: Doudna is not a successor to the El Capitan supercomputer (an AMD/GPU-based system at Lawrence Livermore National Laboratory for classified weapons simulation work). Instead, it succeeds the Perlmutter system at NERSC and focuses on unclassified open science .

What This Means

Dell's ISC 2026 launch positions the PowerEdge XE8812 as the core building block for both commercial AI cloud deployments (beginning with CoreWeave) and the DOE's next open-science supercomputer, Doudna. The Vera Rubin platform represents Dell's flagship HPC and AI infrastructure for the 2026–2027 cycle, powering everything from enterprise AI inference and training to fundamental scientific research in fusion, genomics, and astronomy.