NVIDIA Mellanox® MCX556A-EDAM ConnectX-5 Ex VPI 100GbE/EDR Adapter — PTP/SMA Timing Bracket
Specifications
Manufacturer documentation: NVIDIA ConnectX-5 VPI adapter card user manual: specifications. T.E.S IT-SOLUTIONS is independent of NVIDIA.
Identity
The NVIDIA Mellanox® MCX556A-EDAM is a dual-port ConnectX-5 Ex Virtual Protocol Interconnect (VPI) network adapter (NVIDIA P/N 900-9X5AD-0056-STA, verify against unit label). Each QSFP28 port runs independently at up to 100GbE Ethernet or EDR InfiniBand 100Gb/s on a PCIe 4.0 x16 host interface. The distinguishing feature is the bracket with SMA connector for Precision Time Protocol (PTP) hardware-clock distribution. The MCX556A-EDAM is electrically equivalent to the standard MCX556A-EDAT — only the bracket and timing functionality differ. It is not a Multi-Host adapter.
EDAT vs EDAM — At-a-Glance Comparison
Feature MCX556A-EDAT MCX556A-EDAM ASIC ConnectX-5 Ex ConnectX-5 Ex Ports 2 × QSFP28 VPI 2 × QSFP28 VPI Host interface PCIe 4.0 x16 PCIe 4.0 x16 Bracket Tall, standard Tall with SMA connector PTP hardware timing distribution Not via SMA Yes — PHC over SMA Typical use case General-purpose RDMA / RoCE / VPI Time-synchronized workloads (HFT, 5G, broadcast)
Engineering Detail
PCIe 4.0 x16 provides ~252 Gb/s of effective host bandwidth (after 128b/130b encoding) — sufficient to sustain both 100Gb/s ports concurrently at line rate. The ConnectX-5 Ex ASIC family runs firmware in the 16.x branch. Per-port mode switching is via:
mlxconfig -d <device> set LINK_TYPE_P1=<ETH|IB|AUTO> LINK_TYPE_P2=<ETH|IB|AUTO>
A host reboot is required after the change. Each port is independent; AUTO negotiates from the connected peer.
Timing & PTP Capabilities
PTP Hardware Clock (PHC): on-card hardware timestamping engine for IEEE 1588v2 PTP.
SMA connector: exposes the PHC for PPS (Pulse Per Second) signal distribution to external timing equipment. PPS direction should be verified against the deployed unit; commonly PPS-out on this generation.
Accuracy: sub-microsecond PTP precision per NVIDIA specification when paired with a stable PTP grandmaster.
Linux software stack: linuxptp (ptp4l + phc2sys) on top of MLNX_OFED.
Windows software stack: WinOF-2 with Windows Time Service.
SyncE: not supported on the ConnectX-5 generation. For SyncE in addition to PTP, use ConnectX-6 Dx or ConnectX-7.
What This Is NOT
NOT a Multi-Host adapter. The "M" in this EDAM SKU identifies the SMA timing-bracket variant, NOT Multi-Host. For Multi-Host on this family use the dedicated Multi-Host SKUs (typically with "M" in the model body, e.g., MCX556M-...).
NOT HDR or NDR capable. Max per-port speed is 100Gb/s (EDR / 100GbE). HDR (200Gb/s) requires ConnectX-6; NDR (400Gb/s) requires ConnectX-7.
NOT the non-Ex MCX556A-ECAT. ECAT is PCIe Gen3 and throttles when both ports run concurrently; this Ex card is PCIe Gen4.
NOT SyncE-capable. PTP only on this generation.
NOT bundled with QSFP28 transceivers, DAC cables, or low-profile bracket. Tall (full-height) bracket only.
Networking Offloads & Capabilities
RDMA over Converged Ethernet (RoCE v1/v2) and native InfiniBand verbs.
NVMe over Fabrics (NVMe-oF) — supported via RoCE v2 or InfiniBand transport with appropriate driver and software stack.
Open vSwitch hardware offload via ASAP² (Accelerated Switching and Packet Processing).
IPsec encryption and decryption offload.
T10-DIF (data integrity field) offload for storage workloads.
Erasure coding offload.
SR-IOV virtualization — up to 1,024 VFs total (max 256 VFs per PF, up to 16 PFs).
Deployment & Use Cases
Financial trading: hardware-timestamped market data feeds, PTP-synchronized HFT order paths.
5G fronthaul: ITU-T G.8275.1 PTP profile distribution to radio units.
Broadcast media: SMPTE 2110 IP-based production synchronization.
Industrial / TSN: Time-Sensitive Networking deployments where deterministic latency requires PTP.
HPC clusters: hardware-timestamped collectives, AI/ML training fabrics on EDR InfiniBand.
Storage fabrics: NVMe-oF over RoCE v2, Lustre / LNet, BeeGFS, IBM Spectrum Scale (GPFS).
Virtualization & SDN: SR-IOV with ASAP² hardware Open vSwitch acceleration.
Compatibility & Hard Constraints
Switches (Ethernet): NVIDIA Spectrum / Spectrum-2 100GbE.
Switches (InfiniBand): Switch-IB 2 and other compatible EDR-class NVIDIA / Mellanox InfiniBand switches.
Cabling: QSFP28 DAC, AOC, SR4, LR4. Backward-compatible to FDR / SDR InfiniBand and 1/10/25/40/50 GbE.
Drivers: NVIDIA MLNX_OFED (Linux), in-box mlx5_core (upstream Linux), WinOF-2 (Windows), nmlx5 (VMware ESXi).
Slot: PCIe 4.0 x16 required for full dual-port line rate. PCIe 3.0 x16 throttles aggregate to ~128 Gb/s. Gen4 x8 also throttles.
Bracket: tall bracket only — low-profile not included.
Timing source: SMA functionality requires an external PTP grandmaster or BMC-distributed timing source.
Why T.E.S IT-SOLUTIONS
Every MCX556A-EDAM is checked (see Inspection and checks). Link-up validated at 100 Gb/s per port; PCIe Gen4 link training and dual-port concurrent traffic confirmed. Connector, bracket, and SMA port condition inspected. Professional packing, global B2B shipping from Haifa, Israel.
Frequently Asked Questions
What per-port speed does the MCX556A-EDAM have?
The MCX556A-EDAM per-port speed is 100 GbE Ethernet OR EDR InfiniBand 100 Gb/s (VPI).
What is the MCX556A-EDAM NOT compatible with?
It is not a Multi-Host adapter. Confirm fitment against your platform before ordering MCX556A-EDAM.
What condition is the MCX556A-EDAM supplied in, and who sells it?
The MCX556A-EDAM is supplied in Refurbished condition by T.E.S IT-SOLUTIONS from Haifa, Israel, with 1-3 business day handling and worldwide B2B shipping. See tes-itsolutions.com.