In recent days Intel and Nvidia have introduced or announced new networking products with a common goal of offloading networking traffic to the network processor, thus freeing up the CPU for computational work.
Intel announced a new networking initiative to capitalize on what it calls “a perfect storm of 5G, edge buildout and pervasive artificial intelligence” with an expanded lineup of hardware, software and solutions for network infrastructure.
This includes enhancements to Intel’s software reference architecture, FlexRAN; Intel virtualized radio access network (vRAN) dedicated accelerator; network-optimized next-generation Intel Xeon Scalable and D processors (codenamed “Ice Lake”); and upgraded Intel Select Solutions for Network Function Virtualization Infrastructure (NFVI).
“Implementing a cloud architecture brings the same server economics that transformed the data center and enhances security by allowing networks to be more rapidly repaired, updated and protected. Intel’s full suite of feature-rich silicon, software and tools—and experience in supporting customers transform networks—enable quicker deployment across core, access and edge,” the company said in a statement.
Intel’s networking portfolio includes:
FlexRAN: New optimizations to Intel’s software reference architecture include massive multiple input, multiple output (MIMO) mid-band pipeline for increased bandwidth and support for ultra-reliable low-latency communication (URLLC.)
Intel vRAN Dedicated Accelerator ACC100: a low-power and low-cost acceleration solution for vRAN deployments offloads and accelerates compute-intensive forward error correction from the CPU, thus freeing up the CPU for computational jobs.
Intel is also preparing network-optimized 3rd Generation Intel Xeon Scalable processors, codenamed “Ice Lake-SP”, that are designed for infrastructure use cases that require higher performance per watt, including wireless core, wireless access and network edge workloads and security appliances. Those Xeons will ship at the end of the year.
Intel is also working on a new generation of Intel Xeon D processors, codenamed “Ice Lake-D,” for less performance-sensitive, form factor-constrained environments at the edge and will offer greater levels of integration like built-in networking IP, according to the company. Intel expects to start shipping these processors to customers in mid-2021.
Nvidia DPU plans
As if it hasn’t made enough news lately, Nvidia kicked off its GPU Technology Conference (GTC) with a raft of announcements around graphics and AI, but there was some data center-oriented news as well as the company gave an update on its data processing unit (DPU) initiative.
Nvidia inherited a line of Arm-based high-performance network controllers called BlueField when it purchased Mellanox earlier this year. The company said its controllers introduce the concept of “data-center-infrastructure-on-a-chip architecture,” or DOCA, a mix of Arm processor cores, ConnectX-6 SmartNIC technology, and in some cases, Nvidia’s data center GPU architecture.
“The data center has become the new unit of computing,” said CEO Jen-Hsun Huang in a statement. “DPUs are an essential element of modern and secure accelerated data centers in which CPUs, GPUs and DPUs are able to combine into a single computing unit that’s fully programmable, AI-enabled and can deliver levels of security and compute power not previously possible.”
Like Intel’s RAN processors, BlueField is designed to take over certain network processing tasks from the CPU, to let the CPU do its primary job of computation. A single BlueField-2 DPU can deliver the same data-center services that could consume up to 125 CPU cores, taking a significant load off the CPUs.
BlueField-2, is sampling with customers now and will be available in 2021 in servers from Asus, Atos, Dell Technologies, Fujitsu, Gigabyte, H3C, Inspur, Lenovo, Quanta and Supermicro. A second product, BlueField-2X, features AI features from the new Ampere GPU architecture for AI-accelerated security, network and storage tasks. It will also be available next year.
Beyond that, Nvidia laid out a DPU product road map that includes the BlueField-3 and BlueField-3X in 2022 and the BlueField-4 in 2023 which will integrate the GPU and Arm cores in the silicon. Right bow the BlueField cards come with separate chips for the Arm processor and Nvidia GPU.