Summary
By leveraging Everpure FlashBlade//S R2 with NVIDIA DGX GB200 & GB300 SuperPOD-certified reference architectures, enterprises can build efficient, scalable, and sustainable AI factories.
Enterprises are racing to build their own AI factories—modern production systems designed to turn raw information into intelligence. As NVIDIA defines it, an AI factory requires two fundamental inputs: electricity and data.
These inputs power the immense mathematical operations that produce tokens, the very currency of AI that delivers intelligence and drives business outcomes. While the need for compute and electricity are clear, these factories are quickly realizing that traditional data infrastructure wasn’t designed for the scale, speed, and efficiency AI demands.
To keep pace, enterprises need a solution that delivers the data component in the smallest, most efficient power footprint possible. The new NVIDIA DGX SuperPOD with DGX GB200 or DGX GB300 Systems certified Reference Designs with Everpure® FlashBlade//S™ R2 provide the foundation for this transformation.
This integration brings the AI factory to life, uniting NVIDIA’s most advanced compute platform with the high-throughput, low-latency data access required to keep GPUs continuously fed.
How These Systems Are Redefining AI Infrastructure at Scale
Together, Everpure and NVIDIA enable organizations to build AI infrastructure that’s not only faster and more powerful, but also sustainable, scalable, and ready for the next generation of model training and inference.
The NVIDIA DGX™ GB200 system defines the modern AI Factory with a fully integrated, rack-scale system that unifies up to 72 Blackwell GPUs and 36 Grace CPUs (Arm Neoverse V2) into a single accelerated computing domain. Each rack delivers up to 13.4-13.5TB of HBM3e memory and around 576TB/s of aggregate GPU memory bandwidth, all interconnected through the NVIDIA NVLink™ fabric (NVLink 5th generation), which provides approximately 130TB/s of total GPU-to-GPU bandwidth. The system is liquid-cooled and designed for about 120kW per rack, enabling extreme density and energy efficiency.
The NVIDIA DGX™ GB300 system extends this architecture to the NVIDIA Blackwell Ultra generation, integrating 72 NVIDIA Blackwell Ultra GPUs and 36 Arm-based NVIDIA Grace CPUs in a similar 72-GPU NVLink domain. Each rack provides up to roughly 20TB of HBM3e plus about 17TB of LPDDR5X CPU memory (around 37TB of unified fast memory in total), with an NVLink fabric that again delivers about 130TB/s of aggregate GPU-to-GPU bandwidth.
When integrated with FlashBlade//S R2, the result is a unified, high-throughput data pipeline engineered to match the pace of this compute explosion, while minimizing power, space, and operational overhead. The rack is liquid-cooled and designed for around 120kW of power per rack. For scale-up networking, the NVLink fabric supports total GPU-GPU interconnect bandwidth of ~130TB/s in the NVL72 domain.
Each FlashBlade//S R2 system connects to NVIDIA DGX SuperPOD via 16 400 GbE ports, leveraging RDMA over Converged Ethernet (RoCE) to maximize throughput and efficiency. This standards based enterprise class design eliminates complexity and reduces cost while maintaining enterprise-grade performance and reliability—ideal for organizations looking to scale sustainably.
Architected for Efficiency, Performance, and Simplicity
At the heart of this architecture is Everpure FlashBlade//S R2, the industry’s most efficient and powerful scale-out solution for file and object workloads. Built from the ground up for flash with hardware and software co-designed using Everpure DirectFlash® Modules and the Purity operating system, FlashBlade delivers multi-dimensional performance across all workload types, from massive checkpointing in LLMs to high-resolution computer-vision data sets. It delivers the ability to start your AI environments at any scale and evolve the capacity and performance as your workloads evolve.
Key advantages include:
- Energy efficiency: Up to 50% more performance per watt than the previous generation, dramatically reducing data center power and cooling demands.
- Space efficiency: High-density blades pack multi-petabyte capacity into a compact footprint, cutting rack space requirements across NVIDIA DGX SuperPOD deployments.
- Non-disruptive scalability: Start small and expand capacity or performance independently with data-in-place upgrades, resulting in no downtime and no forklift migrations.
- Flexibility and on-demand consumption: Everpure Evergreen//One™ offers industry-leading flexibility with pay-per-use storage while delivering continuous innovation and non-disruptive upgrades.
- Native multi-protocol support: NFS over RDMA (RoCEv2), S3, and SMB access from the same platform without using gateways and bolt-on architecture unifies training, analytics, and data preparation pipelines.
This combination of sustainability, density, and simplicity allows organizations to train and deploy AI models faster while lowering total cost of ownership (TCO) and environmental footprint.
Fueling AI Innovation across Every Industry
The NVIDIA DGX SuperPOD with DGX GB200 or DGX GB300 systems reference design with Everpure FlashBlade//S R2 empowers enterprises to move from pilot AI projects to full-scale production at unmatched efficiency:
- Healthcare: Powering multimodal training across genomic, imaging, and EHR data with reduced data center energy costs
- Financial services: Supporting real-time analytics on terabytes of market data with dense, low-latency storage
- Media and entertainment: Enabling AI-assisted content creation pipelines that process petabytes of video with minimal footprint
- Manufacturing and energy: Running digital-twin simulations and physics-based AI models with maximum performance-per-watt
Validated to Power Your AI Factory
As enterprises scale to thousands of GPUs, data infrastructure efficiency becomes as critical as raw compute power. Together, Everpure and NVIDIA are redefining what AI infrastructure should be—fast, dense, simple, and sustainable.
The Everpure FlashBlade//S R2 with NVIDIA DGX SuperPOD with DGX GB200 or DGX GB300 systems reference design is a blueprint for the future of AI factories. Delivering the most energy- and space-efficient full-stack AI infrastructure ever built, this solution empowers organizations to push the boundaries of AI while keeping power, cost, and complexity under control.
Learn more at our AI solutions page and download the DGX GB300 SuperPOD Reference Architecture.
How to Power AI at Scale
Explore our video series to learn how Everpure solutions enable effortless scalability, empowering you to turn your AI vision into reality with confidence.
