When announcing new artificial intelligence (AI) capabilities developed by his company earlier this year, NVIDIA (NASDAQ:NVDA) CEO Jensen Huang outlined a new type of processor that will become increasingly important in the years ahead: the data processing unit (DPU).
Data centers are becoming busy digital-economy hubs with massive amounts of data moving into and out of them, and organizations rely on these hubs to power cloud computing tasks key to their operations. To capitalize on the opportunity, NVIDIA has announced the release of new DPU chips. The move will build on NVIDIA's momentum in the data center segment (now its largest business end-market) and help it continuously disrupt an industry dominated by chip giant Intel (NASDAQ:INTC).
CPU, GPU, and DPU basics
Over the years, the CPU (central processing unit) invented by Intel has become the primary chip powering desktop and laptop computers, other mobile devices, and even the movement of data within data centers. GPUs (graphics processing units) were later pioneered by NVIDIA to handle advanced and specialized tasks like video game graphics and, more recently, high-order computing tasks handled in data centers like AI.
But Huang has outlined NVIDIA's vision for the future of computing, built on three types of processors: the CPU for generalized tasks, the GPU as a computing accelerator, and the DPU as a specialized manager of data within data centers to handle cloud computing. To that end, NVIDIA recently unveiled the BlueField-2 DPU and accompanying software development kit for engineers building data center infrastructure and cloud applications.
NVIDIA's new DPUs are already being used by server manufacturers like Dell Technologies (NYSE:DELL) and supported by software infrastructure partners VMware (NYSE:VMW) and IBM's (NYSE:IBM) Red Hat. The DPUs are available for sampling now and are expected to start making their way into data-center hardware systems in 2021, and will help the fast-expanding semiconductor company continue its disruption of the industry.