You may have heard of Apache Arrow, a standardized in-memory format for tabular data, and how it can improve the performance of processing data. Less well-known is that Arrow can also help system developers make the jump from CPU based systems to leveraging accelerated hardware for analytics such as GPUs. Why would you want to do this? Because of something we have termed "The Wall".

Follow along with the description of an holistic "machine" designed to support system-level acceleration from compute to memory, networking, and storage. This talk will connect the dots between adopting Arrow standards and leaping over "The Wall": the difference between how fast machine learning systems can compute and how quickly data processing systems can deliver to them.