NVIDIA is announcing several AI storage advancements at the FMS conference, addressing the growing gap between GPU compute speed and storage throughput. Key announcements include open-sourcing the cuFile APIs (part of GPUDirect Storage) with Google, Intel, Meta, and NVIDIA as maintainers, enabling GPUs to read/write storage directly in microseconds. The Storage-Next initiative brings together 40+ storage and flash vendors to standardize GPU-driven storage behavior. NVIDIA's SCADA framework enables massively parallel GPUs to pull only necessary data directly from storage into high-speed memory. The Vera BlueField-4 STX processor delivers up to 3.21x higher throughput than x86 CPUs in compression/encryption pipelines. Security is built into the architecture via a split-privilege model that keeps user-facing code outside the trusted computing base while a privileged component manages protected storage access.
