VDURA, a high-performance data storage company serving AI factories and GPU cloud providers, announced the general availability of VDURA Data Platform V12. The platform is designed to tackle the storage bottlenecks that can leave costly computing capacity waiting for data.
V12 lets operators serve multiple customers from a shared storage system, automate how storage is provisioned and move data between flash and hard disks as workloads change. VDURA will demonstrate the platform at Ai Everything Abu Dhabi on 6 and 7 October.
AI factories use large clusters of graphics processing units, or GPUs, to train models and run AI services. Those processors need a continuous supply of data. Delays in loading a model, saving a training checkpoint or retrieving data for an inference request can leave computing capacity idle. For providers renting GPU capacity to several customers, storage must also keep each customer’s data and performance requirements separate.
V12 addresses both demands through VDURA’s HYDRA architecture. Operators can assign quality of service, namespaces, encryption keys and network isolation to individual customers within a shared storage environment. This allows a provider to manage separate customer services without building and operating a dedicated storage system for each one.
The platform also introduces REST APIs, Kubernetes CSI support and infrastructure-as-code provisioning. These capabilities let operators create and manage storage through the same automated processes used for other parts of their GPU cloud. As demand grows, they can add capacity and provision new customer environments without relying on manual storage configuration for each deployment.
“Neocloud and AI factory operators told us exactly what they need from storage: keep the GPUs fed, isolate the tenants, automate everything, expand capacity for cold data without a second system, and move that data back to flash the moment it warms up for extended context,” said Ken Claffey, CEO of VDURA. “V12 is that list, shipped.”
AI workloads also change the way data is used. Training may require rapid access to large datasets, while older files can remain on higher-capacity storage until they are needed again. V12’s Context-Aware Tiering moves data between flash and hard disks as access patterns shift, allowing operators to balance performance and capacity within one platform. File and S3 access are available in the same environment, reducing the need to maintain separate copies as data moves through ingest, training, inference and archive workflows.
For workloads that require rapid data movement, V12 includes RDMA data paths and VDURA’s DirectFlow parallel client, designed to transfer data between storage and GPU nodes with limited CPU involvement. Its VeLO metadata engine supports the large volume of file operations generated by AI workloads. The platform also includes snapshots for checkpoints and recovery, encryption at rest and in transit, and self-healing capabilities intended to keep clusters operating as they expand.
V12 is qualified on a configuration using Supermicro Building Block Solutions. The configuration combines Supermicro systems for all-flash and hybrid storage within VDURA’s software-defined architecture. Operators can adjust the balance of flash and hard disk capacity as their workloads evolve while managing storage through the same platform.
Press Release
