According to Yuval Noah Harari, one of the most famous historians living today, mankind has faced 4 major revolutions:
Unlike previous great technological shifts like the personal computer, client servers or even virtualization, the data revolution is more concealed than preceding revolutions. At the epicenter of this shift is the need to catapult the next major technological leap forward where machine learning is the starting point toward true artificial intelligence. To get there, we need to process data, and lots of it.
In Greek mythology Sisyphus, the first king of Corinth, was punished by being forced to endlessly roll a gigantic boulder up a hill only for it to roll back down when it reached the top, leaving him unable to finish his task for eternity. CPUs are engaged in the Sisyphean battle of “pushing data to the end point”. Despite increasing the CPU core frequency or adding more cores, the CPU by itself has become overwhelmed by the growing data processing demands in today’s data centers.
As we enter the Data-Revolution era, the ever-increasing demands for higher computation performance are driving the creation of new data center accelerators and processing units. Previously CPUs and GPUs were the main sources for compute power. The exponential increase in both data volume and complexity have propelled the creation of a data-centric architecture and new type of accelerator unit.
In a data-centric design, the full infrastructure stack – including compute, storage and networking – is architected to speedily and efficiently capture, analyze, classify, manage and archive large volumes of data.
The exponential increase in data volume and complexity has fostered the creation of a new processing unit – the NVIDIA® Mellanox® BlueField® Data Processing Unit or DPU. The BlueField family of DPUs are integrated circuits that include dedicated and programmable in-network computing engines that participate in the application at run time, analyzing the application data as it is being transferred within the data center, or at the edge.
As a complement to the CPU and GPU, the BlueField DPU processes data as it moves in and out of the server in real-time, enabling high-performance fully programmable data processing while freeing up CPU resources. BlueField moves the compute function closer to where the data resides (data-centric architecture), replacing the outdated paradigm of moving data to where the compute is located (compute-centric architecture).
The BlueField DPU integrates the NVIDIA® Mellanox® ConnectX® best-in-class network adapter, encompassing mature and award-winning hardware accelerators with advanced software programmability to deliver diverse software-defined solutions. BlueField’s unique design, with an array of Arm® cores to enable programmability, and DDR4 memory channels – all running at speeds of up to 200Gb/s – enables a wide variety of data center solutions:
The first generation of BlueField DPUs is already deployed in leading high-performance computing, deep learning and cloud data centers to provide new levels of performance, scale and efficiency with improved operational agility. The market is continually finding new frontiers and driving the development of new BlueField generations.
Mellanox is a leading pioneer in data centers with proven in-network computing, network and storage accelerations. The combination of CPUs, GPUs, and DPUs is the next leap in co-processing, which takes advantage of these revolutionary hardware acceleration technologies with software-defined programmability to answer data-centric and edge-computing architecture challenges, maximize return on investment, and make the impossible… possible.
To learn more about BlueField-2 NVIDIA DPU solutions, check out these supporting resources: