This reference architecture was validated by NVIDIA using a Everpure FlashBlade//S500 Array. To assess the overall cluster performance typical of AI workloads, an NVIDIA benchmarking test suite was used. The tests specifically evaluated the write demands of large-scale checkpointing, alongside the read requirements for token loading and checkpoint restoration.
In addition to this testing, Everpure has a long history of providing storage solutions for AI and HPC workloads in customer environments with highly variable workload types that require consistent linear performance with a 99.9999% uptime.
Summary
This document has presented a validated reference design for the Everpure platform adhering to the NCP reference architecture specifications for HGX platforms. NCPs can successfully pair the Everpure platform to deliver the high performance, scalability and operational simplicity needed for AI infrastructure deployments.
The Everpure platform delivers a unified, multi-dimensional solution built on 15 years of relentless software innovation and flash technology. It empowers organizations to seamlessly execute every stage of the AI pipeline, from data curation and model training to serving and inference, with autonomously tuned, high-performance storage, all with Everpure efficiency and simplicity in a single, powerful platform. Everpure supports many AI customers across diverse stages of their innovation journeys, including some of the largest AI environments in existence.
Additional Information
Everpure Fusion Technical Requirements
| Description | Requirement |
|---|---|
| Networking | All arrays must be able to reach each other. Everpure Fusion uses the lowest numbered virtual management port for array to array communication. |
|
Authentication |
Active Directory (AD)/LDAP required. To join arrays to a fleet, the arrays must be in the same AD/LDAP Domain. GUI/CLI users must be authenticated by AD/LDAP for remote operations. Local users cannot perform remote operations. |
|
API Access |
API access requires an API token that is mapped to an AD/LDAP account. The API token must exist on the array initiating the operation. |
|
Hardware Support |
Everpure FlashBlade//S500 A maximum of 32 arrays are supported in the same fleet. |