NVIDIA has made its cuObject client and server libraries generally available. These libraries provide APIs and a wire protocol for faster object storage access using RDMA. RDMA allows data to transfer directly between memory and storage without routing through the server CPU.
The xio-sig consortium is expanding to include cuObject alongside cuFile. Google Cloud is evaluating participation, and Microsoft plans to join the xio-sig board. This move aims to improve storage I/O interoperability across providers.
The new SCADA Server SDK helps storage providers build servers that respond to GPU-initiated requests. IBM Storage has demonstrated a prototype that integrates SCADA with IBM Storage Scale. This allows storage to work more efficiently with GPU workloads.
The NVIDIA Storage-Next initiative coordinates over 40 vendors and customers. It aims to define open standards for GPU-driven, fine-grained storage access. The SCADA software infrastructure supports high-throughput, low-latency data transfers.
Developers can now use cuObject APIs to build applications that access storage directly over RDMA. This reduces data transfer time and CPU use, improving performance for AI tasks. The libraries support both file and object storage protocols.
To get started, review the cuObject documentation and download cuObject Server 2.0.0. These tools help build faster, more efficient AI storage solutions.