Specifications

Overview

The NVIDIA ConnectX-8 SuperNIC is the industry's first 800 Gb/s InfiniBand XDR network adapter, built on PCIe Gen6 x16 for next-generation AI supercomputing clusters. Delivering XDR (4 lanes × 200 Gb/s) with sub-600ns latency and 400M+ messages per second, ConnectX-8 supercharges GPU-to-GPU communication in NVIDIA Blackwell and Rubin architecture deployments.

Available in four configurations: the C8180 (standard-cooled OSFP single-port, XDR InfiniBand / 400GbE) and C8180L (partner-cooled variant) for maximum bandwidth HPC/AI fabrics, plus the C8240 (dual QSFP112, Ethernet-optimized) and C8220 (dual QSFP112, 200GbE value tier). All models support InfiniBand IBTA v1.7a, RoCE v2, GPUDirect RDMA/Storage, and NVMe-oF for converged AI data center networking.

ConnectX-8 introduces an optional auxiliary PCIe Gen5 x16 interface via MCIO harness for multi-host or composable infrastructure. Advanced time synchronization with configurable PPS, time-triggered scheduling, and PTP-based packet pacing enables tightly coupled distributed training at massive scale. Designed exclusively for data center environments with adequate power and airflow — not for desktop or workstation use. QS Compute can source NVIDIA ConnectX-8 SuperNICs for AI cluster deployments across APAC.