Nvidia BlueField DPU

Main articles:

• Processor
• CPU

BlueField - Data Processing Unit (DPU) line. Nvidia Initially, BlueField was the brand of Mellanox, which in 2019 became part of Nvidia.

2021

Integration with Palo Alto Networks

In July 2021, Nvidia announced the integration of DPU BlueField into Palo Alto Networks solutions. This step will allow the technology developer to expand the direction associated with data processing devices (DPU, Data Processing Unit), which appeared in the NVIDIA portfolio after the acquisition of Mellanox. Customers with NVIDIA data center solutions can dramatically improve virtual firewall performance.

NVIDIA BlueField-2 DPU

According to the announcement of Palo Alto Networks, the next-generation virtual firewall, which is in development, will scan network traffic five times faster than solutions running only on CPU, thanks to the use of DPU Nvidia BlueField. The term DPU was invented by NVIDIA in 2020. DPUs have already been used in cloud services, supercomputers, and many OEMs and software developers.

DPU tasks include accelerating virtual network functions of the data center while protecting and optimizing network traffic. Hardware acceleration of data center architecture components becomes fundamental as IT data center infrastructure burdens and applications become more complex. The DPU includes dedicated accelerator functions for various applications, such as:

• hardware acceleration of IPsec, TLS, AES-GCM, Public Key and RegEx for security tools;
• NVMe-oF, RAID, and GPUDirect for storage
• RDMA, RoCE, SR-IOV, VXLAN, VirtIO, and GPUDirect RDMA for network functions;
• hardware acceleration of streaming video and other delay-sensitive traffic.

NVIDIA BlueField-3 DPU

Such infrastructure loads grow faster than their processing capabilities, and already consume about 30% of the capacity of server CPU. DPUs have special hardware accelerators that allow you to perform these tasks more efficiently, freeing up CPU cores for data center applications.

DPUs provide virtualization and security without compromising physical server performance. They range from cloud computing and media streaming to storage, peripherals, and high-performance computing.

NVIDIA CEO Jensen Huang describes DPU as "one of the foundations of the data center of the future. The CPU is designed for general-purpose computing operations, the GPU is designed to accelerate parallel computing, and the DPU is designed to accelerate I/O and data processing. "

DPUs BlueField provide a secure, hardware-accelerated infrastructure for all workloads in any environment, from cloud data centers to edge computing clusters. DPU ensures that your network, storage, and cybersecurity services are offloaded, accelerated, and isolated. DPU BlueField combines performance computing, infrastructure-on-a-chip programmability, and high-speed networking solutions for the most demanding workloads.

Programming and configuration of the DPU is performed by connecting existing software to already known high-level software interfaces (APIs) of the device. These APIs are included in a DPU software stack called NVIDIA DOCA. It includes drivers, libraries, tools, documentation, sample applications, and runtime for installing, deploying, and orchestrating services on thousands of DPUs installed in the cloud data center.

Muninder Singh Sambi, Senior Vice President of Products at Palo Alto Networks, notes that NGFW's VM-Series firewall acceleration with DPU will provide businesses and telecommunications companies "flexibility and cloud automation without compromising performance. NVIDIA is BlueField well suited for cybersecurity solutions running in cloud environments.

In the data center market, NVIDIA does not plan to focus on accelerating computing and graphics tasks using graphics processors GPUs. The company is confident that it can take on an even greater share of hardware and software to solve the emerging class of problems. This should be helped by DPU BlueField. This type of processor is also planned to be developed due to synergy with the capabilities artificial intelligence and reduced CPU utilization.

Asus, Dell Technologies, Supermicro and other OEMs will introduce certified servers with DPU BlueField-2 at the end of 2021. But for servers with NVIDIA DPU acceleration to succeed, you need to integrate with various key software solutions in network infrastructures. This integration and improved software performance will help explain the need to invest in DPUs that cost more than standard NICs. And it was here that Palo Alto Networks came to the rescue.

The VM-Series NGFW solutions from Palo Alto Networks, according to NVIDIA, will programmatically determine the part of network traffic that should be checked by the firewall, since up to 80% of network traffic is not of interest or cannot be processed by advanced firewall security features. Since the traffic splitting function is performed using the DPU-based Intelligent Traffic Overload service, it is possible to significantly reduce the number of traffic packets sent to the CPU-based firewall.

Traffic splitting using NVIDIA BlueField-2 DPU-based Intelligent Traffic Overload service

As a result, the Intelligent Traffic Overhead service can increase the effective firewall bandwidth to 100 Gb/s with NVIDIA DPU BlueField-2, which is five times higher than the 16 Gb/s bandwidth that is achievable when using processing exclusively by the server CPU. NVIDIA notes that using DPU can also reduce capital expenditures by 150% compared to an outdated approach.

Palo Alto Networks reported that a DPU-accelerated solution could "change the rules of the game" for cloud environments.

According to experts, a joint solution from NVIDIA and Palo Alto Networks is likely to be interesting to large enterprises whose workloads are already highly virtualized.

NVIDIA BlueField DPUs will support more than a dozen partner software solutions, including:

• VMware, along with Project Monterey, provides DPU capabilities to more than 300,000 organizations that leverage VMware solutions with speed, fault tolerance, and security.
• Red Hat is preparing a developer kit for Red Hat Enterprise Linux and Red Hat OpenShift, which are used by 95% of Fortune 500 companies.
• Canonical, in, Ubuntu Linux the most popular operating system among. public clouds
• Check Point Software Technologies uses DPUs in products that use over 100,000 organizations worldwide to prevent cyber attacks.
• Other partners that can be accelerated with DPU include Cloudflare, DDN, Excelero, F5, Fortinet, Guardicore, Juniper Networks, NetApp, Last Data, and WekaIO.

Market analysts at Dell'Oro Group predict that the number of smart network ports delivered will almost double from 4.4 million in 2020 to 7.4 million by 2025.

In preparation for such growth, NVIDIA introduced a product plan for the release of the next two generations of DPUs. BlueField-3, which will begin delivery in 2022, will be equipped with network interfaces with a speed of up to 400 Gb/s and will perform hardware acceleration functions comparable in performance to 300 x86 cores. BlueField-4 will provide an order of magnitude greater performance coupled with NVIDIA's AI computing technologies.

The BlueField-2 series, which entered the market in 2021, comes in various configurations with port speeds from 25 to 200 Gb/s.

Integration with Nvidia Aerial A100 AI-on-5G

On June 28, 2021, NVIDIA announced the expansion of Arm-based CPU support for the NVIDIA Aerial A100 AI-on-5G platform, providing a wider choice of solutions in the 5G ecosystem.

According to the company, the NVIDIA Aerial A100 AI-on-5G computing platform uses the NVIDIA Aerial software development kit and will include 16 Arm Cortex-A78 processors in the NVIDIA BlueField-3 A100. This standalone converged card will provide access to enterprise AI peripheral applications via 5G vRAN with Watt optimized performance and shorter deployment times. More details here.

BlueField-3 The A100 can be deployed to both x86 and Arm processors. BlueField-3 is a next-generation data processing device designed for AI and computational optimization. It is optimized for 5G connections, multi-user clouds, and offers software-defined, hardware-accelerated communications, storage, security, and management services on the periphery. The BlueField-3 A100 is expected to be available in the first half of 2022.

BlueField-3

On April 12, at the opening of its annual GTC conference, NVIDIA announced DPU BlueField-3, which is scheduled to become available in 2022. According to the developer, BlueField-3 will be the first network card with a transfer rate of 400 Gb/s to the port over Ethernet and NDR InfiniBand. In addition, 16 ARM Cortex cores will be available, which will accelerate network computing, thanks to special hardware solutions. NVIDIA states that such a chip replaces about 300 traditional CPU cores. The chip is connected via PCI Express 5.0 and has its own DDR5 memory.

The full list of announcements on GTC 2021 is available here.

2020: BlueField-2

In October 2020, Nvidia introduced two new solutions: BlueField-2X DPU and BlueField-2 DPU. In fact, they are network cards with a fast ARM processor, which takes over the computing load, freeing up the resources of the server's central processor for other tasks.

BlueField-2X combines Arm processor cores, graphics accelerator and Mellanox network adapter on one PCB, and BlueField-2 does not have a GPU on board.