HPE has released version 4 of its Alletra Storage MP X10000 platform, now generally available. The update doubles single-cluster scale to roughly 23 PB of raw capacity and adds native file support alongside object storage on the same system, targeting larger AI training, inference, and data lake consolidation.
Scale and unified protocols
Release 4 supports up to 16 nodes and 16 JBOFs in a single cluster, delivering roughly twice the performance and capacity of the prior release. File and object now operate natively on one platform through native-namespace NFS, removing the need for separate silos or protocol translation overhead. HPE said this gives customers headroom to consolidate AI training, inference, analytics, and large unstructured data repositories without disruptive upgrades.
Deeper NVIDIA integration and inference acceleration
The platform extends RDMA acceleration from object to file workloads and adds GPUDirect Storage enablement, with readiness for NVIDIA AI Enterprise certifications. The X10000 was already the first object storage platform to achieve NVIDIA-Certified Storage validation. HPE said independent testing of the RDMA-accelerated X10000 for KV cache offload "showed up to 20x faster time to first token and up to 17x higher effective inference throughput, improving GPU utilization and lowering cost per token."
Data intelligence and security controls
Release 4 advances the X10000 Data Intelligence platform with NVIDIA NIM-based multimodal AI capabilities, multinode data intelligence deployments, and support for customer-developed AI functions. The goal is a shorter path from raw data to AI-ready data without standing up a separate preparation stack. For security-conscious environments, HPE added expanded support for disconnected and air-gapped deployments, opening conversations with government and regulated industries. Subscription options now include 1-, 6-, and 7-year terms.
Why this matters for IT and development professionals
For teams managing infrastructure that feeds GPU clusters, the combination of doubled scale, native file and object unification, and RDMA-accelerated access directly affects utilization and cost per token. The move to collapse silos onto one platform also reduces operational overhead when preparing unstructured data for inference and agentic workflows. IT managers evaluating storage for AI pipelines can assess how these capabilities align with existing NVIDIA environments and data governance requirements - a topic covered in depth through AI Strategy for IT Managers learning paths. For technology leaders shaping infrastructure roadmaps, the shift toward unified, air-gap-ready platforms carries architectural implications explored in AI Technology Leadership Courses.
Your membership also unlocks:
