Rethinking Standard NFS for AI and HPC workloads: A Validated Architecture from Xinnor, Supermicro and AMD

Date: Tuesday, May 12, 2026

Time: 5:00 PM CET / 8:00 AM PST

AI workloads are pushing storage demands not only for the traditional HPC but as well for the enterprise AI-infrastructure. High-performance NFS over RDMA is part of that shift. AI-class storage is no longer just a specialist requirement, more enterprise teams need it now. At the same time memory and NVMe remain constrained, which keeps many conventional high-performance storage systems expensive and hard to source. In this webinar Xinnor, Supermicro and AMD will walk through xiNAS, a shared storage platform built to bring that level of performance to a wider set of teams. It runs on Supermicro servers with AMD EPYC processors and has been validated for AI and HPC workloads. The goal is simple: high throughput, strong resilience, and simpler operations than most HPC-style storage stacks.

Joint validation on Supermicro AS-1116CS-TN systems showed up to 74.5 GB/s read and 39.5 GB/s write from a single RAID6 protected server. In a two-server configuration, the system reached 117 GB/s read and 79.6 GB/s write. We will break down the design choices behind those numbers and explain why a well-tuned NFS stack can be a practical shared-storage option for modern AI and HPC deployments.

We will also cover the operational side. xiNAS is built to reduce infrastructure overhead, avoid proprietary client dependencies, and make deployment, tuning, and lifecycle management easier. With the xiNAS Installation Kit, client-side automation tools, built-in management workflows, HealthCheck monitoring and optimization, and MCP support teams can deploy and operate the system through a largely automated setup. The session will also show how MCP and agent-driven operations can help manage shared storage in modern environments.

Watch the recording