Data for AI
SOPA Image/LightRocket (via Getty Images)
Cloudian, a longtime leader in scalable object storage technology, directly addresses the complexity of AI storage infrastructure with the new Cloudian HyperStore with Nvidia GPUDirect for object storage.
Announced at Supercomputing Conference SC24 in Atlanta, Cloudian brings together scale and performance in a single solution, enabling AI storage environments with a unified data lake that serves as a shared data repository for all stages of the AI lifecycle. Simplify.
This new product is the industry’s first object storage solution that leverages Nvidia GPUDirect technology to reduce latency and increase throughput to meet the high demands of AI training and other demanding GPU-centric workloads.
AI creates new demands on storage
AI workflows rely on vast amounts of unstructured data, creating new and unprecedented data management challenges for IT organizations. Adding to the challenge of scale is the complexity of managing a myriad of storage requirements, from training data to checkpoint data, vector databases, log files, and more.
Traditional storage solutions are designed for the more modest demands of traditional enterprise applications and struggle to meet these demands. Addressing these complexity, cost, and scaling limitations is critical to providing efficient and cost-effective data management for high-performance AI applications.
Object storage for AI
Object storage is already the data management foundation for hyperscale cloud providers and offers a clear solution to AI data challenges. Just as it revolutionized cloud infrastructure, object storage will transform AI data management by efficiently processing massive unstructured datasets while maintaining simplicity.
Object storage brings multiple benefits to AI workflows.
- Scalability: Object storage systems can scale almost infinitely with a flat address space and support multiple data types in a shared data lake.
- cost efficiency: Leveraging standard server technology and operating at scale, object storage provides a cost-effective solution for storing and managing the multi-petabyte datasets needed to train AI models.
- safety: Object storage provides comprehensive security for AI workflows through multiple integrated protections. Beyond data encryption at rest and in transit, we can provide data immutability for ransomware protection, multi-tenancy, and granular access controls to manage permissions across your AI team.
- S3 compatibility: Native S3 API compatibility and cloud integration simplifies data return from the cloud and ensures interoperability with many AI tools that employ S3 APIs.
- Metadata management: Object storage enables rich, customizable metadata that facilitates efficient tagging, searching, and management of data, which is essential for organizing and retrieving data for AI model training and inference.
Cloudian’s on-premises object storage solution, HyperStore, is a scalable, secure, S3-compatible platform designed to manage massive amounts of unstructured data. HyperStore’s exabyte-level scalability allows organizations to start small and scale seamlessly as their data requirements grow. Adding GPUDirect technology to your solution improves support for AI and other high-performance GPU-centric workloads.
What is Nvidia GPUDirect?
Nvidia’s GPUDirect technology is at the heart of Cloudian’s new HyperStore solution. It enables high-speed data transfer between storage and GPU memory by providing a direct data path that bypasses the traditional CPU route. This reduces the number of steps and avoids CPU and system memory bottlenecks, resulting in faster data access.
Nvidia GPU Direcet
NAND research
Hyperstore GPUDirect for object storage
Cloudian’s new HyperStore and Nvidia GPUDirect for Object Storage take high-performance data management a step forward. GPUDirect’s seamless integration with HyperStore allows users to benefit from Cloudian’s scalability and ease of use while enabling a high-performance data path dedicated to GPU workloads.
Cloudian solutions have a compelling performance story. Cloudian reports continuous throughput speeds of 200 GB/s from the HyperStore system using GPUDirect. This is three times the throughput using GPUDirect compared to traditional data access methods measured on the same all-flash system. This high-performance throughput makes data available to the GPU at the speeds needed for demanding workloads.
By delivering file system performance directly from object storage, GPUDirect eliminates the need for a separate file storage layer, simplifying architecture while reducing infrastructure costs.
It also increases efficiency. Cloudian says GPUDirect frees up processing power and reduces CPU usage by an estimated 42%. This makes it easier for organizations to manage large-scale data without increasing hardware footprint or operational costs, improving performance and contributing to lower power consumption and lower costs in the long run.
Cloudian’s GPUDirect solution flattens your storage architecture into a single data lake, eliminating complex data migrations while simplifying the management of large unstructured datasets. For AI workflows, this means training data, models, and inference can all leverage the same high-performance storage platform, accelerating development while ensuring data consistency.
There is also a security story, as GPUDirect technology operates without the need for vendor-specific kernel-level changes, reducing exposure to vulnerabilities and improving system security. This is valuable in fields such as healthcare and finance, where data security and regulatory compliance are paramount.
Analyst’s view
Rapid adoption of AI will result in data growth, requiring efficient, scalable, and high-performance storage infrastructure. Cloudian HyperStore and Nvidia GPUDirect for Object Storage meet this demand by providing advanced platforms optimized for AI and GPU-centric workloads that require seamless scale data management.
Cloudian transforms traditional AI storage architectures by consolidating unstructured data into a single, high-performance data lake. Native S3 API and cloud integration facilitate seamless data management between cloud and on-premises environments.
Nvidia GPUDirect support for object storage completes the picture, enabling direct object-to-GPU transfer, eliminating the complexity and cost of separate file and object storage tiers. The result is dramatically simpler infrastructure, improved performance, and lower operating costs.
Cloudian isn’t the only storage solution on the market that utilizes Nvidia GPUDirect. Storage solutions from Dell Technologies, DDN, WEKA, and VAST Data all support this technology. However, Cloudian is the first technology provider to integrate GPUDirect directly into objects. Leverage storage solutions to stay ahead of data trends for AI and other GPU-based workloads.
Nvidia GPUs are at the heart of AI-driven enterprise data transformation and are critical components at nearly every stage of the AI lifecycle. Cloudian’s new GPUDirect object storage solution ensures that enterprises have the tools to harness the full potential of their data. It helps businesses across industries leverage data to achieve more while controlling costs and improving efficiency, a key capability for delivering AI-driven digital transformation. An attractive solution.
Disclosure: Steve McDowell is an industry analyst and NAND Research is an industry analyst firm that engages in research, analysis, and advisory services with many technology companies, including all companies mentioned in this article except DDN. doing or being engaged in. The companies mentioned were not involved in the drafting or publication of this article. Mr. McDowell has no equity interest in any of the companies mentioned.
