Nvidia Opens Up GPU Storage Access with Open-Source cuFile API

Nvidia has released its cuFile application programming interfaces (APIs) and the storage software stack beneath them as open source. This move is part of Nvidia's broader push into AI storage through an initiative called Storage-Next.

The cuFile APIs allow GPUs to read from and write to storage directly, bypassing CPUs and reducing latency. According to Nvidia, this can cut storage access times to microseconds using GPU threads and high-bandwidth memory.

Nvidia is also promoting a framework called SCADA (scaled, accelerated data access) that splits direct data access into two parts to address security risks. This allows user-level application components to remain outside the trusted computing base while establishing protected access between the application and approved storage using standard Linux security protocols.