The term SmartNIC refers to a type of network interface card (NIC), usually a circuit board or chip, that acts as a programmable accelerator. While SmartNIC is the most common name for these processors, the nomenclature varies by manufacturer – with some calling the products data processing units (DPUs) or distributed services cards (DSCs).
SmartNICs offer a flexible, hardware-based solution to accelerate processing across workloads and can help reduce the cost of hardware expenditure in the data center.
How does a SmartNIC work?
A network interface card (NIC) is a hardware component that connects computers on large scale or local area networks (LANs) by converting data packages into signals which then spread throughout the network. NICs are most often comprised of configurable silicon units. SmartNICs differ from the traditional NIC in that they can be programmed to facilitate sophisticated packet processing functions and contain the intelligence to offload functions from the host’s CPU.
The majority of SmartNICs are highly specialized hardware units called accelerators. They are able to handle communication workloads more efficiently than CPUs, but equally important is their ability to be programed and reprogrammed to handle evolving network needs.
SmartNICs also boast integrated multi-core processors. High-performance SmartNICs can be equipped with up to 100/200/400 GbE ports, a set of programmable acceleration engines and even a GPU to support artificial intelligence (AI) workloads. They are also relatively easy to integrate into data center servers using the PCIe slot on most high-performance motherboards.
Why are SmartNICs important?
SmartNICs are important for a number of reasons, the first being that traditional NICs are struggling to keep up with the ever-increasing demand for cloud services. As virtualization and microservices continue to proliferate in the cloud, alongside increasing mobile phone and cloud service provider use, workloads are growing faster than traditional data center CPUs can handle. SmartNICs address the growth by offloading workloads and freeing up CPU usage, which is especially vital to the cloud as data centers continue to innovate to meet processing demand.
“The data center is the new unit of computing,” said NVIDIA CEO Jensen Huang at the NVIDIA GTC 2021 Keynote. “A third of the roughly 30 million servers shipped each year are consumed by running the data center’s software-defined networking stack. This workload is increasing much faster than Moore’s law, so unless we offload and accelerate this workload the data center will have fewer and fewer CPUs to run applications."
SmartNICs are able to address this issue directly, at the network edge. They can also be upgraded in firmware to be able to address future issues, offering versatility in terms of modifiable functions.
Leading cloud providers are already using custom SmartNIC designs in their data centers, including Azure with Azure SmartNIC and Amazon Web Services with AWS Nitro. In fact, according to an Intel survey and report only 2% of respondent service providers have no plans to deploy SmartNICs in the near future.
Chip manufacturers are also piling into the space, including Pensando, NVIDIA, Netronome, Marvell, Intel and Broadcom. Available products include NIVIDIA’s BlueField, Pensando DSC or the Broadcom Stingray.
What are key SmartNIC applications?
SmartNICs offer a variety of features:
- Networking functions – SmartNICs are capable of a range of networking functions including network address translation (NAT), firewalling, load balancing, overlay networks, routing and telemetry.
- Storage controllers – SmartNICs can also be used as storage controllers, managing both the hard drives (HDDs) and solid-state drives (SSDs) of the data centers server.
- CPU offloading – Equipment manufacturers commonly use SmartNIC multi-core processors to offload the workload of CPUs in data centers, this allows users to optimize performance from their existing hardware as opposed to replacing it.
- DDOS security – F5, an American technology company specializing in application security and multi-cloud management, claims utilizing SmartNIC technology has allowed them to develop a security solution capable of withstanding a DDOS attack 300x in magnitude than what is covered by current software solutions, according to an article by Tom Atkins, an F5 product marketing manager.
What are the limitations of SmartNICs?
There is no such thing as a perfect solution and SmartNICs are not an exception to that rule. SmartNICs also have an array of limitations that are important to understand for potential investors.
- No standard – As a new technology, there is not a regulatory authority that can define and enforce what a SmartNIC is and what can be called a SmartNIC. This can lead to confusion among customers that may think they are paying for one thing and manufacturers that are delivering something else.
- Reduced troubleshooting – Because SmartNICs are a hardware solution, there is a significant lack of troubleshooting solutions that often leave DevOps teams with nothing else to suggest in the instance of system failure, aside from restarting the equipment.
- High cost – While NICs are relatively inexpensive to produce and utilize, the complexity of SmartNICs means their price ranges from three to up to 10 times the cost of a standard NIC.
- Power consumption – As several hundred virtual machines run within a given NIC, implementing a SmartNIC can be a very complex and drawn-out process that results in a hugely compute- and power-intensive solution.
Learn more about SmartNIC
Achronix pushes the boundaries of networking with 400 GbE and PCIe Gen 5.0 for SmartNICs
Smart NIC market to grow at a 42 percent CAGR through 2027, According to Dell'Oro Group
Read more of our cloud explainers here.