Network Card Mellanox Technologies MCX555A-ECAT

The Mellanox Technologies MCX555A-ECAT is an intelligent RDMA-enabled network adapter card designed for high-performance computing, cloud, Web2.0 and storage platforms. This internal network card provides two 100 Gb/s ports supporting both InfiniBand and Ethernet connectivity, a PCI Express host interface, sub-600 ns latency, very high message rates, and advanced offload capabilities including PCIe switch and NVMe over Fabrics support.

Advantages

MCX555A-ECAT offers low latency and high bandwidth suitable for data‑intensive workloads, with hardware offloads for MPI, SHMEM/PGAS, Rendezvous Tag Matching and support for out-of-order RDMA operations. The adapter supports both IBTA RDMA and RoCE, provides multipath reliability via switch adaptive routing and ordered completion semantics even when data is delivered out-of-order, and includes Burst Buffer offload and Dynamic Connected Transport for scalable checkpointing and transport efficiency.

Suitable for

• High-Performance Computing clusters and MPI-based workloads
• Machine learning training and inference environments
• Data analytics and large-scale distributed computing
• Cloud and Web2.0 services requiring low-latency networking
• Storage platforms using NVMe over Fabrics and burst buffering

Key specifications

• Brand: MELLANOX
• Model / MPN: MCX555A-ECAT
• EAN: 7290108480313
• Interface: PCI Express (host interface)
• Connectivity: 2 ports, InfiniBand and Ethernet (VPI)
• Maximum data rate: 100000 Mbit/s per port (100 Gb/s)
• Latency: sub-600 ns
• Features: RDMA (IBTA) and RoCE, MPI/SHMEM offload, Rendezvous Tag Matching, NVMe over Fabrics offload, PCIe and Network Atomic operations

How to use

Install the MCX555A-ECAT into a compatible PCIe slot on the host server according to the server manufacturer’s installation guidelines. Connect fiber or compatible cables to the card’s ports and configure the network settings in the host operating system using vendor drivers and tools provided by Mellanox. For RDMA and NVMe over Fabrics deployments, enable the appropriate firmware and software stacks (RDMA drivers, OFED or vendor-supplied packages) and configure routing and offload parameters as required by your cluster or storage architecture.

Recommendations

Use vendor-recommended firmware and drivers, validate cabling and link speeds, and test MPI/RDMA performance in a staging environment before production deployment. For optimal throughput and latency, combine the card with a compatible switch that supports adaptive routing and the chosen RDMA protocol (InfiniBand or RoCE), and ensure host CPU, PCIe lanes and storage configurations match the intended workload.