Intel, hardware vendors working on a high-performance network card

Intel announced a collaboration with Inspur, Ruijie Networks, and Silicom Connectivity Solutions to design and develop new infrastructure processing units (IPU) using both a CPU and FPGA.

IPU—what Intel calls a data-processing unit (DPU)—is a programmable networking device designed to offload network processing tasks such as storage virtualization, network virtualization, and security from the CPU. That reduces overhead and frees up the CPU to focus on its primary data-processing functions. They are becoming a real growth industry, with multiple products on the market from Nvidia, Marvell, Fungible, and Xilinx.

Intel has already introduced one IPU, codenamed Mount Evans, and is moving on to a next-generation product called Oak Springs Canyon or officially the C6000X. It has both a Xeon D processor and Intel Agilex FPGA to handle the networking with custom programmable logic. Both processors have 16GB of DDR4 memory.

More importantly, there are two 100Gbps ports along with a PCIe Gen4 x16 connectivity. It can support cloud-infrastructure workloads such as Open vSwitch, NVMe over Fabrics (NVMEoF), and Remote Direct Memory Access (RDMA) over Converged ethernet v2 (RoCEv2). Like previous IPU cards, the C6000X is made by Silicom, not Intel.

Intel has previously stated that its vision of programmable networks is based upon open standards where the IPU is integral to delivering new network capabilities, such as telemetry, congestion control, and traffic steering.

China is a pretty big market for FPGAs, which is why the two Chinese firms Inspur (cloud infrastructure) and Ruijie Networks (digitization of all industries) are involved.  Both will deploy Oak Springs Canyon to deliver customized IPU-optimized solutions.