Limited Time Offer: Save $350 on your Ai Cluster → Shop Now

--

Days

--

Hours

--

Mins

--

Secs

Sale Ended

TECHNOLOGY AT A GLANCE

Explore the capabilities

2 × 400G QSFP112 ports

Connect compatible network fabrics at up to 400Gb/s per port, with up to 800Gb/s total adapter bandwidth.

PCIe 6.0 x16 host interface

High-bandwidth PCIe connectivity supports demanding server and accelerator architectures.

RDMA and RoCEv2 acceleration

Reduce CPU overhead and support efficient data movement across compatible InfiniBand and Ethernet networks.

AI-ready congestion control

Advanced, programmable congestion control helps maintain efficient traffic flow in large-scale GPU and data center fabrics.

High-speed networking for AI at scale

ConnectX-8 brings 400G-per-port connectivity directly into the server, helping AI, HPC, and cloud environments reduce communication bottlenecks between GPUs, storage, and the fabric. It is designed for infrastructure where network performance is as critical as compute performance.

TECHNICAL DETAILS

Built into the system

Dual-port 400G SuperNIC

The C8240 provides two QSFP112 network ports, each capable of up to 400Gb/s connectivity. Together, the adapter delivers up to 800Gb/s of total bandwidth for compatible high-performance server and accelerator environments.

InfiniBand and Ethernet networking

ConnectX-8 supports both InfiniBand and Ethernet networking. Compatible InfiniBand deployments support speeds up to 400Gb/s per port, while Ethernet deployments support 25Gb/s, 50Gb/s, 100Gb/s, 200Gb/s, and 400Gb/s connections.

PCIe 6.0 x16 host interface

A PCIe Gen6 x16 interface provides the high-bandwidth host connection required for modern GPU servers, storage nodes, and data-intensive workloads. This C8240 card uses a half-height, half-length form factor with a tall bracket.

RDMA and RoCEv2 acceleration

RDMA and RoCEv2 capabilities help enable efficient, low-overhead data movement across compatible network fabrics. GPUDirect RDMA support allows compatible GPUs and network adapters to exchange data more directly, reducing unnecessary CPU involvement.

Advanced traffic control for AI fabrics

ConnectX-8 supports programmable congestion control, traffic management, and telemetry-driven capabilities designed for large AI and HPC fabrics. These capabilities help manage communication-heavy workloads such as distributed training and high-performance storage access.

Cloud, virtualization, and security offloads

The adapter supports Ethernet Accelerated Switching and Packet Processing (ASAP2) for compatible SDN and virtualization environments, including overlay and virtual-switch acceleration. Secure boot, secure firmware update, flash encryption, and device attestation support platform security requirements.