Joerg Hiller
Nov 14, 2025 12:10
NVIDIA’s RDMA technology improves AI storage efficiency by enhancing S3-compatible storage solutions, offering higher throughput and lower latencies for AI workloads.
In a significant advancement for AI storage solutions, NVIDIA has unveiled its RDMA (Remote Direct Memory Access) technology tailored for S3-compatible storage. This innovation is set to enhance the performance and efficiency of AI workloads, according to a report from NVIDIA’s official blog.
Revolutionizing AI Workloads
As AI demands continue to grow, the need for scalable and cost-effective storage solutions has become paramount. By 2028, enterprises are expected to generate nearly 400 zettabytes of data annually. This increase, coupled with the predominance of unstructured data such as audio, video, and images, calls for innovative storage solutions. NVIDIA’s RDMA technology addresses these challenges by accelerating the S3-API-based storage protocol, optimized specifically for AI data.
Benefits of RDMA for S3-Compatible Storage
The RDMA for S3-compatible storage offers multiple advantages over traditional storage methods. It provides higher throughput per terabyte, reduced latency, and lower costs, making it an attractive option for AI applications. Additionally, it offers:
- Cost Efficiency: Reduced storage costs can facilitate faster project approval and implementation.
- Portability: AI workloads can seamlessly operate across on-premises and cloud environments, using a common storage API.
- Enhanced Performance: Faster data access benefits AI training, vector databases, and key-value cache storage for AI inference.
- Reduced CPU Usage: RDMA offloads data transfer from the host CPU, freeing resources for AI processing.
Industry Adoption and Collaboration
NVIDIA is collaborating with industry leaders to standardize and adopt RDMA for S3-compatible storage. Companies like Cloudian, Dell Technologies, and HPE are integrating this technology into their high-performance storage products. Cloudian’s HyperStore, Dell’s ObjectScale, and HPE’s Alletra Storage MP X10000 are some of the solutions benefiting from RDMA enhancements.
According to Jon Toor, CMO at Cloudian, the standardization efforts with NVIDIA are set to bring scalability and performance improvements to S3-based applications. Similarly, Rajesh Rajaraman from Dell Technologies highlights the unmatched scalability and performance RDMA brings to their ObjectScale storage solution. HPE’s Jim O’Dorisio also emphasized the reduced latency and cost benefits of integrating RDMA technology into their storage offerings.
NVIDIA’s RDMA libraries are initially available to select partners, with a broader release expected through the NVIDIA CUDA Toolkit in January. As the technology gains traction, NVIDIA’s Object Storage Certification will further facilitate the integration of RDMA solutions across the industry.
For more information, visit the NVIDIA blog.
Image source: Shutterstock