Xinnor Introduces xiNAS, High-Performance NFS over RDMA Storage Optimized for AI and HPC
Validated on Supermicro AS-1116CS-TN NVMe servers powered by AMD EPYC processors
This is a Press Release edited by StorageNewsletter.com on February 13, 2026 at 2:00 pmXinnor Ltd., developer of high-performance storage software, announced xiNAS, a high-performance NFS storage solution purpose-built for AI, HPC, and other data-intensive workloads.
Validated on Supermicro AS-1116CS-TN NVMe servers powered by AMD EPYC processors, xiNAS delivers all-flash, scale-out performance using standard NFS semantics, without proprietary clients or specialized hardware.
Modern AI training, simulation, and analytics pipelines demand massive shared throughput with predictable latency and uninterrupted operation during failures. Traditional NAS architectures often become a bottleneck in these environments. xiNAS changes that equation by combining Xinnor’s xiRAID software RAID engine with an optimized XFS filesystem and NFS over RDMA, maximizing the performance of the underlined hardware.
In joint validation testing with Supermicro, xiNAS demonstrated up to 74.5GB/s read and 39.5GB/s write throughput from a single server equipped with just 12 NVMe PCIe Gen5 drives, and up to 117GB/s read and 79.6GB/s write throughput in a 2-node scale-out configuration, demonstrating linear scalability, while maintaining strong performance during drive failures and rebuilds. Backend testing showed 97–100% efficiency of theoretical NVMe performance with minimal CPU overhead, preserving headroom for high-speed networking.
“xiNAS is designed to make NFS a performance enabler rather than a constraint for AI and HPC,” said Dmitry Livshits, CEO, Xinnor. “By tightly integrating xiRAID with an optimized filesystem and RDMA-enabled NFS, we deliver shared storage that keeps GPUs fed, scales linearly, and continues to perform even under real-world fault conditions.”
xiNAS is optimized for multi-client, multi-server deployments and supports linear bandwidth scaling as nodes are added.
Key benefits include:
- Extreme throughput using standard NFS, accelerated with RDMA (RoCE or InfiniBand)
- No proprietary client software, enabling simple deployment and integration
- High efficiency data protection with xiRAID, delivering strong performance in both healthy and degraded modes
- Operational stability during failures, with minimal performance impact during drive failures and rebuilds
The validated configuration used Supermicro AS-1116CS-TN servers equipped with AMD EPYC 9004-series processors, NVIDIA BlueField-3 DPUs, and PCIe Gen5 NVMe storage, highlighting xiNAS’s ability to fully exploit modern CPU, network, and storage capabilities.
“Supermicro continues to focus on delivering application-optimized building blocks for AI and HPC,” said Lawrence Lam, VP, AI and storage solutions, Supermicro, Inc.. “Our NVMe-optimized AS-1116CS-TN servers, combined with Xinnor’s xiNAS software stack, demonstrate how standard architectures can deliver exceptional shared-storage performance for the most demanding workloads. By combining a high-performance storage engine with an optimally pre-configured NAS server platform, xiNAS significantly simplifies and accelerates deployment for AI initiatives in vertical markets like scientific laboratories, universities, and hospitals.”
“AI and HPC workloads place intense demands on both compute and storage,” said Derek Dicker, CVP, enterprise business group, AMD (Advanced Micro Devices, Inc.). “AMD EPYC processors provide the core density, memory bandwidth, and I/O capabilities required to support high-performance, RDMA-enabled storage solutions like xiNAS, helping customers build balanced platforms that keep accelerators and CPUs operating at high utilization. xiNAS is engineered to efficiently leverage AMD EPYC CPU capabilities to sustain extraordinary levels of data access performance, even in worst-case conditions such as component failures or sudden peak workload bursts, ensuring consistent service levels when they matter the most.”
xiNAS is available through Supermicro and its partners. The solution is delivered as complete software and hardware combo, enabling customers to easily deploy high-performance and resilient NFS storage.
Resources:
Blog: xiNAS: High-Performance NFS Storage for AI & HPC on Supermicro AS-1116CS-TN Servers
Validated configuration are available in the solution brief
Comments
Xinnor has entered the file storage space with a new solution - an unexpected move that coincides with the departure of Davide Villa, who has left the company to join ThinkParq. ThinkParq also offered a file storage solution, being a partner of Xinnor, focusing on high-end requirements for HPC and AI environments with its parallel file system, BeeGFS.
Xinnor had already shared promising research results at the MSST conference last September, presenting compelling performance data in a file storage context. At that time, Xinnor did not officially offer a file storage product, but those early results suggest the groundwork for this solution was already underway.
Click to enlarge
The newly introduced offering is an NFS-based solution combining Xinnor XiRAID, NVMe-oF, and NFS with RDMA support. It leverages tuned XFS file system parameters along with multiple RDMA- and NFS-optimized settings, including daemon thread tuning (with 64 threads proving effective), nconnect and max_connect, session trunking, no-wdelay, READ_PLUS operation, LOCALIO, and NFS Inter-Server Copy. The Xinnor presentation delivered at MSST provides further technical detail on this architecture.
In addition, Xinnor achieved strong recognition in the most recent IO500, ranking third overall when integrated with Lustre, as published at ISC25.
Subscribe to our free daily newsletter
