Nvidia Open-Sources cuFile API for Faster GPU Storage

Nvidia Open-Sources cuFile API for Faster GPU Storage
Nvidia is open-sourcing its cuFile API, enabling millisecond-speed direct data access between storage and GPU memory. The technology bypasses the CPU and system memory using direct memory access, eliminating GPU starvation bottlenecks that slow AI workloads. Alongside this, Nvidia launched Storage-Next, an industry initiative with 40 vendors including Kioxia and Micron, built around its SCADA framework. SCADA enables massively parallel GPU data access while maintaining security by separating privileged and unprivileged computing components to prevent data conflicts.
Read the original article →