Placă de rețea Mellanox Technologies MCX556A-ECAT

The Mellanox Technologies MCX556A-ECAT is an internal network adapter designed for high-performance computing, data center, cloud and storage environments. This ConnectX-5 based card provides two 100 Gb/s ports with InfiniBand and Ethernet support, low latency (sub-600 ns) and high message-rate capability. It includes PCIe host connectivity and advanced offloads such as NVMe over Fabrics and PCIe switch offload to improve throughput and reduce CPU overhead.

Advantages

The MCX556A-ECAT delivers dual 100Gb/s connectivity with low latency and very high message rates, enabling demanding HPC, analytics and cloud workloads. It supports MPI offloads (including Tag Matching and AlltoAll optimizations), advanced dynamic routing and in-network compute capabilities to accelerate data-path algorithms. ASAP2 accelerated switching and packet processing reduce virtual switch CPU load while improving data transfer efficiency. An alternative Socket Direct configuration allows use in servers without a full x16 PCIe slot by splitting the 16-lane bus into two x8 connections for dual-socket direct CPU access.

Suitable for

• High Performance Computing (HPC) clusters and MPI-based applications
• Cloud and NFV platforms requiring RDMA/RoCE acceleration
• Data analytics and high-throughput storage systems including NVMe-oF
• Web 2.0 and large-scale data center network environments

Key specifications

• Model: Mellanox MCX556A-ECAT (ConnectX-5)
• Ports: 2 × 100 Gb/s (InfiniBand and Ethernet)
• Latency: sub-600 nanoseconds (typical)
• Host interface: PCI Express (16 lanes, split-able to 2× x8 for Socket Direct)
• Offloads: NVMe over Fabrics, PCIe switch offload, MPI offloads, ASAP2 accelerated switching/packet processing
• Use cases: RDMA, RoCE, in-network compute, virtual switch offload

How to use

Install the MCX556A-ECAT into a compatible PCIe slot on the host server following the server and card manufacturer instructions. When using the Socket Direct variant, connect the auxiliary PCIe connection card and dedicated harness as specified to provide two x8 connections. Configure appropriate drivers and firmware from Mellanox, enable RDMA/RoCE or InfiniBand stack as required, and tune host network settings and MPI or storage software to make use of available offloads (MPI offloads, NVMe-oF, ASAP2). Verify link, firmware versions and network fabric settings after installation.

Recommendations

For best results in HPC and latency-sensitive workloads, use up-to-date Mellanox drivers and firmware, enable relevant offloads (RDMA/RoCE, MPI Offloads) in software stacks, and validate cabling and switch compatibility for 100Gb/s links. In dual-socket servers consider the Socket Direct configuration to provide each CPU with a dedicated x8 path to the adapter.