Paper Title: DPUs Help Systems Process Huge Data

Paper Abstract: A Data Processing Unit (DPU) is a programmable processor focused on dealing with large amounts of data at very high speed. It is an essential addition to systems that handle today's huge problems in real-time analytics, AI/ML, HPC, media processing, and cybersecurity. The DPU acts as a third socket in the data center, alongside the CPU and GPU, with the purpose of providing infrastructure services for the software-defined datacenter. DPUs provide a high-speed fabric between compute and storage servers, allowing processing to move to the edge and avoiding system-saturating moves of large datasets. The result is much higher bandwidth and much lower cost than standard processors can provide. An example is cloud-to-edge intelligent processing, which requires a powerful specialized processor to avoid tying up central compute elements and bogging down systems. Such processors can utilize a many-core architecture that combines advanced data handling features with the functions of a SmartNIC, a smart storage adapter, and an AI accelerator.

Paper Author: Guru Bachchu, Lead Applications Engineer, Kalray

Author Bio: Adam Tarnowski is Sr VP/General Manager Americas for Kalray, where he is in charge of all marketing, business development, sales, and engineering functions. He has a long history of successful engagements with large clouds and other major customers in promoting semiconductor and services sales, business development, and marketing. He has led both AMD and Flex in producing huge sales increases year-to-year. He was also co-founder and Director of Sales for Hyve Solutions, a producer of rack-based systems for hyperscalers. He has previous experience with both Intel and Freescale.