The Dell AI Factory with NVIDIA

The Dell AI Factory with NVIDIA introduces a new modular architecture designed to help enterprises adopt and scale AI workloads seamlessly, from pilot projects to full-scale production. At its core, the solution offers flexible starting points, from the Small Inferencing, ideal for piloting AI use cases and moving them to production, and the Enterprise Fine-Tuning and Training Foundation, which supports larger-scale AI training. The modular design allows organizations to start small and scale incrementally with expansion modules as business needs evolve. Combined with the Dell Automation Platform, and built on open platform technologies like Ubuntu and Red Hat OpenShift, the Dell AI Factory with NVIDIA simplifies deployment and management, offering pre-validated blueprints for rapid AI workload rollouts. Achieve faster results, reduce complexity, and drive innovation at scale.

The Dell AI Factory with NVIDIA is powered by multiple Dell PowerEdge server models tailored to different AI workloads: the PowerEdge XE9680, a flagship 8‑GPU system ideal for training and fine‑tuning with NVIDIA H100 Tensor Core GPUs; the PowerEdge R760xa, a 2U GPU‑accelerated workhorse commonly used for L40S‑based inferencing and fine‑tuning as well as general GenAI tasks; the PowerEdge R660, often used in validated stacks for supporting/management compute and lighter AI services; and the PowerEdge XE7745, featured in Spectrum‑X solution IDs for building large‑scale GPU clusters. Together with NVIDIA accelerators, BlueField DPUs, and high‑performance Spectrum‑X networking, these systems deliver a full‑stack platform for enterprise GenAI from PoC to production.

The Blackwell-based Dell PowerEdge XE9780 improves LLM training by combining dual Intel Xeon 6 CPUs (up to 86 cores) with eight NVIDIA HGX Blackwell (B200/B300) GPUs in a 10U air cooled chassis purpose built for GenAI training and fine tuning. In practice, this delivers up to 4x faster training versus prior platforms, accelerating time to value and reducing iteration cycles for model development. Its air cooled, standard rack design simplifies integration into existing data centers, enabling you to scale performance without costly site upgrades. Backed by Dell’s enterprise management and support, XE9780 provides a reliable, production ready foundation for LLM training and fine tuning initiatives.

The default stack includes NVIDIA AI Enterprise (enterprise AI platform and support), NVIDIA NIM microservices (optimized model endpoints), and optional NVIDIA Omniverse for simulation workflows—validated and sold through Dell as part of the AI Factory solution.

Yes, the Dell AI Factory fully supports advanced cooling technologies, including direct-to-chip liquid cooling. This is essential for managing the thermal output of high-density GPU configurations, such as those using the NVIDIA Blackwell platform. Implementing liquid cooling allows you to deploy more compute power per rack, improve your data center's power usage effectiveness (PUE), and reduce overall operational expenses, ensuring your infrastructure is both powerful and efficient.

The Dell AI Factory, supports a maximum of 72 NVIDIA Blackwell GPUs in a single, liquid-cooled rack. This remarkable density is achieved through the NVIDIA NVL72 system, which integrates the GPUs with high-speed NVLink interconnects.

The Dell AI Factory with NVIDIA is designed to accelerate your time-to-value by dramatically reducing the deployment time and complexity associated with do-it-yourself (DIY) AI infrastructure. Instead of spending months integrating disparate hardware and software, you can deploy a fully validated and optimized solution.

You can virtualize H100/H200 with NVIDIA vGPU (via NVIDIA AI Enterprise) to share GPUs across VMs, but MLPerf results are typically published on bare metal configurations; for highest benchmark targets, use pass through or bare metal rather than vGPU.

Direct-to-chip liquid cooling, a core feature of the Dell AI Factory's high-density configurations, significantly lowers your data center's Power Usage Effectiveness (PUE) and operational expenditures (OPEX). By transferring heat directly from the GPUs with liquid, it removes thermal constraints far more efficiently than traditional air cooling.