Making a GPU into a Data Access Engine

"Cj Newburn (Distinguished Engineer) - Nvidia Vikram Sharma Mailthody (Nvidia) - Nvidia A new generation of applications need to satisfy demand for huge volumes of fine-grained accesses from 100K GPU threads more than they need compute power. Datasets are growing too big to fit into memory. A new programming model is needed- and a new generation of storage needs to be defined. NVIDIA is proposing that new programming model for GPU-initiated storage IO: SCADA- for Scaled Accelerated Data Access. NVIDIA is also creating an ecosystem of vendors to define requirements for a next generation of storage that is focused on IOPS/$ and on optimizing tail latency. Come participate in defining the future of AI storage access!"