The IT Nerd

VDURA today announced the general availability of the VDURA® Data Platform V12, the release that turns the platform into a multi-tenant, API-driven storage service for GPU clouds and AI factories. V12 ships as a qualified solution on Supermicro Building Block Solutions®, scaling from 8 to 100,000 GPUs on one software stack. VDURA will showcase V12 at Ai Everything Abu Dhabi, 6–7 October at ADNEC Centre, Booth H3-D45.

Neoclouds and AI factories exist to rent and run GPUs, and storage determines how much of that GPU capacity turns into revenue. Every accelerator waiting on a checkpoint, a model load, or a cold read is margin lost. Every tenant that cannot be isolated is a contract that cannot be signed. Every kilowatt spent on storage is a kilowatt not spent on compute. V12 was engineered for that operating model: keep every GPU fed, keep every tenant isolated, keep every cluster online, and do it on hardware operators already standardize on.

What is new in V12

V12 extends the HYDRA architecture, VDURA’s High-performance, Yield-optimized, Distributed, Resilient Architecture, with the capabilities multi-tenant GPU infrastructure requires:

• Multi-tenant by design. Per-tenant quality of service, namespaces, encryption keys and VLAN isolation on one shared fleet, so a provider can carve a single storage pool into hard-walled tenant services with capacity and performance guarantees.
• API-first automation. REST APIs, Kubernetes CSI and infrastructure-as-code tenant provisioning, so storage is deployed, provisioned and billed through the same pipelines as the rest of the GPU cloud.
• Context-Aware Tiering™. Data lands on the right media automatically as access patterns shift between training and inference. Roughly 90% of files stay on flash while roughly 90% of capacity settles on HDD, in one platform with no stub files, no rehydration steps and no manual tuning.
• Persistent context for inference. A KV cache that outlives the pod. Sessions resume instead of prefilling again, delivering faster first tokens and more concurrent users on the same GPUs, at flash cost rather than recompute cost. 
• RDMA data paths. Direct GPU-to-storage transfers with the CPU out of the path. The DirectFlow™ parallel client takes roughly 191 MB of DRAM and zero cores from the GPU node.
• Elastic Metadata Engine. VeLO™ metadata acceleration of up to 20x improvement, 225,000 creates and deletes per second per Director, and billions of metadata operations per second in aggregate.
• File and S3 in one platform. An S3 object is a file in the volume, not a copy of one. No staging copies between ingest, training, inference and archive.
• Snapshots and SMR HDD optimization. Instantaneous, space-efficient snapshots for checkpoints and operational recovery, and SMR HDD unlocking 25 to 30% more capacity per rack.
• End-to-end encryption. AES-256 at rest and in flight, with KMIP key management per tenant.
• Self-healing resiliency and VDURA Sentinel™. Failure domains as small as a single VPOD™, no manual rebuilds and no downtime windows, backed by VDURA Sentinel^TM proactive support that opens the service request, with the diagnosis attached, before the customer sees a fault.

Qualified on Supermicro

V12 is qualified on 100% Supermicro Building Block Solutions. The qualified configuration uses the Supermicro AS-1116CS-TN, a 1U system with a single AMD EPYC™ 9005 series processor and 12 NVMe bays, as both the VeLO Director node and the all-flash F-Node, alongside the AS-2015HS-TNR hybrid storage node and the CSE-947HE2C 4U 90-bay JBOD for the mixed-fleet data plane. Every node connects with RDMA straight to the GPU nodes and no dedicated back-end storage fabric.

Built for GPU economics

Clusters grow online from three nodes to thousands, and flash share is a dial rather than a fork. A single platform at roughly 20 PB usable spans a 35x performance range, from a capacity-optimized mixed fleet at 2% flash and 18 kW to an all-flash configuration at 1,000 MB/s per TB, so operators size storage to the workload instead of standing up a second system when the workload changes. The result is 2x+ performance per watt and more than 60% lower total cost of ownership than competitive architectures at the same feed rate, which returns power, rack space and capital to the operator for more GPUs.

Proven at scale

V12 builds on 25 years of parallel file system engineering, from PanFS to VDURA, with more than 1,000 production deployments in over 50 countries and namespaces of more than 1,500 nodes. VDURA Data Platform V12 was named AI Data Management Solution of the Year in the 2026 AI Breakthrough Awards.

Meet VDURA and Thor at Ai Everything Abu Dhabi

VDURA will demonstrate V12 on Supermicro at Ai Everything Abu Dhabi, 6–7 October 2026, ADNEC Centre, Booth H3-D45. VDURA partner Hafþór “Thor” Björnsson (The Mountain from Game of Thrones and Strongman), whose record-setting data lifts have powered VDURA, will be at the booth alongside Team VDURA. Operators building or expanding GPU capacity in the region can book time with the team at AI Everything Abu Dhabi.

Availability

VDURA Data Platform V12 is generally available today for all V5000 class systems and as an upgrade for V11 customers. VDURA Sentinel^TM is included with V12; proactive service requests and parts dispatch are delivered through VDURACare Premier, VDURA’s 10-year support offering covering hardware, software and 24×7 expert response under a single contract. Learn more at vdura.com/data-platform.

VDURA today announced VDURA Sentinel™, a proactive support capability included with the VDURA® Data Platform V12. Sentinel runs natively on the platform, watching continuously for faults and other notable events. Customers can share that health data with VDURA’s support team through the Sentinel cloud service, where it is analyzed against patterns drawn from thousands of VDURA systems in the field. When Sentinel spots an issue, it automatically opens a service request with VDURA support, complete with a proposed fix, often before anyone on site knows there’s a problem.

For operators running GPU clusters, the cost of a storage fault is measured in the hours between the fault and the fix: idle accelerators, stalled training jobs, and for a neocloud, tenant SLAs at risk. Conventional storage support adds delay at every step. Someone notices a symptom, files a ticket, gathers logs, describes the problem, and waits for a technician to work out what happened. Sentinel removes that cycle.

Sentinel continuously reads health, capacity, and alert telemetry from each system and compares it against failure patterns VDURA has resolved across thousands of systems in the field. When a pattern matches, the VDURA support team receives a plain-language summary, a trend view, and a recommended action, with the service request already open and, where needed, a parts dispatch initiated on the customer’s behalf. The first person to look at the ticket is looking at a diagnosis, not a blank page.

Sentinel is built on the same API-first foundation as the rest of the V12 platform. The collector gathers its telemetry through the platform’s published APIs, and everything Sentinel sees is equally available to the customer’s own dashboards and observability tooling. There is no privileged vendor-only view: operators get fleet-level observability into their storage as a first-class part of the platform, and VDURA support sees only the subset the customer chooses to share.

Built for security-conscious infrastructure teams

AI factories and GPU cloud providers operate under strict data-handling and tenant-isolation requirements, and many will not accept a vendor agent with write access or an opaque telemetry stream. Sentinel was designed around those constraints:

Read-only, outbound-only. The collector runs inside the customer’s environment with read-only credentials and makes only outbound HTTPS connections. There is no inbound access and no ability to change configuration or touch data.

File data never leaves the building. Sentinel does not transmit file contents, credentials, user or workload information, or logs. Only four categories of operational metadata can be shared: fleet health and capacity metrics, node inventory, alerts, and deep-dive diagnostics.

Per-category control, no support call required. Each of the four categories can be switched off independently, or all sharing can be disabled entirely, by the customer at any time. The local dashboard and the underlying storage are unaffected either way.

Inspect before it ships. A built-in payload inspector shows the exact contents of every outbound push before it is sent.

Auditable code. VDURA publishes the collector as dependency-free, readable Python so customer security teams can review exactly what it does.

Availability

Sentinel is included with the VDURA Data Platform V12 release, including the local fleet health dashboard and API access to the underlying telemetry. Proactive service requests and parts dispatch are delivered through VDURACare Premier, VDURA’s 10-year support offering that covers hardware, software, and 24×7 expert response under a single contract. 

Learn more: vdura.com/vdura-sentinel-support

VDURA and Wasabi Technologies, the hot cloud storage company, have announced a technology alliance to connect GPU-adjacent AI data infrastructure with predictably priced S3-compatible cloud storage. 

Through the alliance, the companies will collaborate to help AI factories, neoclouds and enterprise HPC environments place active datasets and checkpoints close to GPUs while extending long-term retention, protection and data reuse into the cloud.

The alliance is anchored by a simple operating principle: keep AI data close when it is being used and move it freely when it is not. AI infrastructure teams routinely keep far more inactive data on performance infrastructure than they need to, because moving it has meant operational complexity, access delays or unpredictable cloud costs. The result is performance capacity consumed by data that no longer requires it, and retained data that is harder to protect, govern and reuse than it should be. 

VDURA’s high-performance data storage serves the active phase of the AI data lifecycle: dataset staging, model loading, training, checkpointing and inference. VDURA provides GPU-adjacent parallel file system storage performance with RDMA data paths and full POSIX workflows, spanning NVMe flash, HDD capacity and a native S3 interface within a single global namespace.

Wasabi extends the environment with S3-compatible cloud object storage serving as an active archive and long-term retention layer. Datasets, checkpoints, model versions and derived artifacts that are no longer in active use move to Wasabi for protection, compliance, disaster recovery and distribution to other sites or compute environments, and remain readily accessible for retraining, model comparison and governance. Wasabi’s predictable pricing, with no per-GB egress or API request charges under its standard terms, allows teams to plan retention and reuse costs in advance rather than discovering them on an invoice.

Tiered data placement is a practice long proven by the largest cloud operators, who pair lean performance infrastructure with deep, low-cost capacity. The alliance brings a comparable design pattern to organizations building their own AI infrastructure, using two independent, specialized platforms connected through open interfaces rather than a single closed stack.

The joint technology alliance will be present at Ai4 2026, August 4 to 6 at The Venetian in Las Vegas, where Wasabi will exhibit at booth #935. Ai4 is North America’s largest artificial intelligence industry event, gathering more than 12,000 attendees across enterprise AI, infrastructure and research.

VDURA today announced three major advances for ISC High-Performance 2026: a next-generation multi-tenant control plane, significant S3 performance improvements and native S3 object tagging. Together, these capabilities are designed to simplify operations and accelerate data-intensive workloads at scale for AI and HPC organizations worldwide.

Planned for general availability in the second half of 2026, these advances continue VDURA’s mission to deliver the world’s most powerful data infrastructure platform for AI and HPC.

Next-Generation Control Plane: Simplified, Multi-Tenant Management for AI and HPC Environments

VDURA’s next-generation control plane, planned for general availability later this year, is purpose-built for the operational demands of modern AI and HPC environments. It introduces a modern management interface that delivers a streamlined tenant administration model, enabling platform and tenant operators to manage complex multi-tenant storage environments from an intuitive dashboard. Alongside the new UI, a REST API for key platform operations gives operators and integrators a foundation for automation and tooling workflows. Together, they significantly reduce operational overhead while preserving the depth of control that enterprise and research organizations require.

S3 Performance Improvements: Sustained Throughput for Cloud-Native AI Pipelines

VDURA is delivering targeted S3 performance improvements designed to sustain high throughput across cloud-native AI pipelines, including AI model checkpointing, inference serving, and large-scale dataset ingestion. These improvements reduce latency for S3-native operations and increase aggregate throughput for concurrent read and write workloads, ensuring the VDURA Data Platform continues to deliver the performance AI operators demand as workloads scale.

S3 Tags: Metadata-Driven Control Across the Data Lifecycle

Building on enhanced S3 performance, VDURA is also introducing native S3 object tagging support, enabling organizations to attach rich metadata to objects stored on the VDURA platform. S3 Tags unlock policy-based data lifecycle management, automated tiering workflows, and fine-grained access controls across AI training datasets, model artifacts, and scientific data collections. For HPC and AI operators managing petabyte-scale object stores, S3 Tags provide the metadata infrastructure required to enforce governance policies and manage data across its full lifecycle.

Availability

The Next-Generation Control Plane, S3 Performance Improvements, and S3 Tags are planned for general availability in the second half of 2026 for all V5000 class systems. Existing customers can upgrade in place via an online software update. Customers interested in early access are encouraged to visit VDURA at ISC 2026 or contact the VDURA sales team at vdura.com.

VDURA at ISC High Performance 2026

VDURA is exhibiting at ISC High Performance 2026 in Hamburg, Germany. Visit the VDURA booth G21 to experience these new capabilities firsthand and learn more about VDURA’s full portfolio of AI and HPC storage solutions.

 VDURA today announced Version 12 of the VDURA Data Platform, a major software release that redefines scalability and resilience for AI and high-performance computing (HPC) environments. V12 increases performance on every metric and introduces a new elastic Metadata Engine, system-wide Snapshot Support, and optimized integration for hyperscale class HDD’s using Shingled Magnetic Recording (SMR) technology to deliver unprecedented performance efficiency, durability, and capacity scaling across the data lifecycle.

Building on the performance foundation established in V11, VDURA V12 Key Highlights of Data Platform V12 include:

• Elastic Metadata Engine: A next-generation, scale-out, metadata engine designed for increased performance. It dynamically scales across nodes to sustain performance as file and object counts grow, eliminating traditional bottlenecks and improving metadata operations by up to 20x.
• Snapshot Support: Enables instantaneous, space-efficient point-in-time copies of datasets for AI pipelines, model checkpoints, and operational recovery. Snapshots integrate seamlessly with support for both manual and policy-based retention.
• SMR HDD Optimization: Extends the VDURA data platform to support Shingled Magnetic Recording (SMR) drives for ultra-dense data storage. A new write-placement engine organizes sequential zones intelligently, unlocking 25–30% more capacity per rack without compromising throughput or data integrity. 

Breakthrough Performance Meets Resilient Scale

The Elastic Metadata Engine enables metadata operations to scale linearly across multiple metadata nodes, supporting billions of files and objects under active use. Combined with snapshot and SMR tier support, V12 further reduces total cost of ownership while extending VDURA’s lead in AI-scale storage economics. With aggregate throughput performance increased by more than 20%, V12 delivers unmatched efficiency for training pipelines, inference workloads, and hyperscale data movement.

Availability

VDURA Data Platform V12 will be generally available in Q2 2026 for all V5000 and V7000-class systems. Existing customers running V11 or later can upgrade in place via an online software update with zero downtime.

The Mountain vs. the Data Transfer Record at SC25

On Nov 18, 2025, at 11:30 AM CT, Hafþór “Thor” Björnsson returns to SC25 Booth #2033 with VDURA to break the AI and HPC data transfer record. Powered by Phison Pascari SSDs and the VDURA Data Platform, Thor will move hundreds of petabytes in seconds, demonstrating the next level of data performance and scalability.

Last year, Thor stunned the crowd with his World Record Data Lift, hoisting over 282 PB of capacity, the most data ever lifted. This year, it’s all about speed.

VDURA has announced the launch of its first scalable AMD Instinct™ GPU reference architecture in collaboration with AMD. The new validated blueprint defines how compute, storage and networking should be configured for efficient, repeatable large-scale GPU implementations. The design combines the VDURA V5000 storage platform with AMD Instinct™ MI300 Series Accelerators to eliminate performance bottlenecks and simplify deployment for the most demanding AI and high-performance computing (HPC) environments. 

AI and HPC pipelines are increasingly limited by storage that cannot keep pace with growing data volumes. This slows GPU utilization, increases energy costs and reduces overall efficiency. The new reference architecture is engineered to keep AMD Instinct™ GPUs fully utilized, delivering sustained performance with a design that is efficient, expandable and simple to operate. 

Following a technical evaluation, AMD selected VDURA for its AMD Instinct™ GPU-optimized performance, low client overhead and proven ability to scale. The solution has already been chosen for a U.S. federal systems integrator AI supercluster, demonstrating its readiness for mission-critical workloads. 

The reference architecture provides compute, storage and networking at scale. It supports 256 AMD Instinct™ GPUs per scalable unit, achieves throughput of up to 1.4 TB/s and 45 million IOPS in an all-flash layout, and delivers around 5 PB of usable capacity in a 3 Director and 6 V5000 node configuration. Data durability is assured through multi-level erasure coding, while networking options include dual-plane 400 GbE and optional NDR/NDR200 InfiniBand. 

Built to grow with demand, the modular design allows organizations to add Director Nodes for extra performance, expand with all-flash storage for more bandwidth, or combine flash and HDD capacity for cost-effective growth, all within a single namespace.  

For further information, the full 20-page architecture reference guide is available for download at: https://www.vdura.com/wp-content/uploads/2025/07/VDURA_AMD_AI_Reference-Architecture-2.pdf