The NVIDIA ConnectX-8 SuperNIC, as a core component of high-performance networking solutions, relies on a compatible ecosystem of cables, optical modules, and switches to enable high-bandwidth, low-latency connectivity in data centers and supercomputing environments. This document outlines the key parameters of the ConnectX-8 SuperNIC, typical application connection schemes, and introduces the optical modules and fiber optic patch cords used alongside it.
ConnectX-8 SuperNIC Overview
The NVIDIA ConnectX-8 SuperNIC is a high-performance network adapter designed specifically for hyperscale AI computing. It incorporates advanced routing and telemetry-based congestion control technologies, significantly improving network efficiency and AI workload performance. Its InfiniBand version further extends SHARP in-network computing capabilities, supporting both large-scale training and inference scenarios while optimizing power consumption, enabling sustainable million-GPU-scale AI data centers.
ConnectX-8 SuperNIC Key Parameters
-
Protocols: Ethernet, InfiniBand
-
Maximum Bandwidth: 800Gb/s
-
Port Configuration and Speed: Single port up to 800Gb/s, or dual ports each up to 400Gb/s
-
Optical Interface Form Factors: OSFP, QSFP112
Typical Connection Schemes and Application Examples
① In the QM3400 switch, dual-port 1.6T OSFP 2DR4 optical modules connect via two direct 8-fiber MPO/PAC fiber optic patch cords to single-port 800G OSFP DR4 optical modules on two ConnectX-8 network adapters.
② In the QM3200 switch, dual-port 1.6T OSFP 2DR4 optical modules connect via two 8-fiber 1-to-2 MPO/APC fiber optic patch cords to single-port 400G OSFP RHS optical modules on four ConnectX-8 network adapters.
TARLUZ Related Products
