Dispatch
NVIDIA open-sources cuFile and introduces SCADA to let GPUs pull only the bytes they need from storage
At FMS, NVIDIA open-sourced its cuFile APIs so GPUs — not just CPUs — read and write storage directly in microseconds, and introduced SCADA (scaled, accelerated data access), a framework letting massively parallel GPUs pull only application-necessary data from storage into high-bandwidth memory. It also showed the Vera CPU in the Vera BlueField-4 STX storage processor delivering up to 3.21x higher throughput than an x86 CPU on a two-stage compression-and-encryption pipeline. Over 40 storage and flash vendors are in the Storage-Next initiative, with DDN integrating SCADA into Infinia; the driver is KV-cache and long-context working sets outgrowing system memory.
Source
↳ Follow the thread