Executive Overview
This reference architecture for an AI Factory with Everpure® FlashBlade® and NVIDIA is built to provide enterprise organizations with design guidance for scale-out storage optimized for AI infrastructure and applications. This solution delivers performance with operational efficiency and simplicity for AI applications running on NVIDIA accelerated computing platforms in either PCIe-optimized or HGX reference configurations.
Everpure and NVIDIA enable enterprises to build AI factories capable of supporting advanced AI workloads with the performance and efficiency required for today's demanding applications.
This document presents validated reference configurations for building AI factories for the enterprise organizations with NVIDIA, integrating Everpure FlashBlade//S500 array systems and Pure1® management software with NVIDIA HGX™ H100, H200, and B200; and NVIDIA RTX™ PRO 6000 Blackwell Server Edition, H100 NVL, H200 NVL, and L40S GPUs.
The data ingestion, processing, and storage requirements for large-scale enterprise AI require the high-performance, highly efficient storage that Everpure FlashBlade was architected to deliver.
-
Scalability: It can scale up to 1024 of NVIDIA GPUs and multiple PBs of DirectFlash®, making it suitable for large-scale AI model training and inference.
-
Performance: It has multi-dimensional performance that scales non-disruptively and ensures the consistent, low latency, and high bandwidth performance that dynamic AI factory workloads need to address the most challenging computational problems of today.
-
Software integration: The solution comes with NVIDIA AI Enterprise software, a cloud-native suite of software tools, libraries, and frameworks, including NVIDIA NIM and NeMo microservices, that accelerate and simplify the development, deployment, and scaling of AI applications.
Solution Overview
Running an AI factory with NVIDIA and Everpure provides key benefits including:
-
Data-in-place, non-disruptive upgrades: Non-disruptively upgrade storage processor and/or storage capacity resources independently to either scale them within a technology generation or upgrade them to newer generations. With FlashBlade, customers can enjoy continuous data-in-place hardware and software innovation without requiring planned downtime and disruption windows.
-
Ease of use and management: Single integrated solution for all stages of an AI data pipeline with a simple architecture, making it easy to deploy, manage, and use.
-
Unmatched efficiency: Extreme performance density enables organizations to maximize GPU utilization and tokens/watt while reducing infrastructure overhead, space, and total costs.
-
Performance at any scale: Start your AI journey at any scale and seamlessly grow and adapt as needed.
-
Agile AI platform: Built for future innovations from Everpure and NVIDIA as the AI landscape and customer requirements evolve.
-
Trusted industry leaders: Everpure and NVIDIA are leaders in the AI market, joining to deliver state-of-the-art, flash-based AI-ready infrastructure since 2018, with the introduction of the Everpure AIRI® solution.1
This platform handles a multitude of use cases in training and inference like generative AI, computer vision, deep learning recommendation systems, high-performance computing, and more.