NVIDIA Mellanox ConnectX-7 MCX75343AMC-NEAC OCP3.0

What is the single-port speed and form factor of this adapter?

The MCX75343AMC-NEAC features a single QSFP port delivering 400Gb/s throughput in the OCP 3.0 Small Form Factor design, offering the highest single-port speed option in the NVIDIA ConnectX-7 series, ideal for hyperscale data centers requiring maximum single-port bandwidth in dense deployments.

What is the factory default protocol mode, and does it also support Ethernet?

This adapter ships with a factory default of InfiniBand NDR 400Gb/s mode. The same port can also be reconfigured to run 400GbE Ethernet mode, letting system integrators use a single SKU to flexibly serve either InfiniBand or Ethernet deployment needs.

Does this adapter support NVIDIA Multi-Host and Socket Direct technologies?

Yes, this adapter supports NVIDIA Multi-Host technology, allowing up to 4 independent servers to share the network ports and management of a single card, and also supports Socket Direct, which splits the card's PCIe x16 connection across two processor sockets to reduce cross-socket latency. Combined with the OCP 3.0 form factor, both features balance hardware consolidation efficiency with compute performance in hyperscale data centers.

What security features does this adapter offer, and does the NEAC SKU include encryption?

The MCX75343AMC-NEAC integrates hardware-based Secure Boot with a built-in crypto engine for embedded IPsec/TLS offload, protecting adapter firmware integrity while reducing the host CPU's cryptographic workload — well suited to financial, telecom, and enterprise cloud deployments with strict data transmission security requirements.

When configured in InfiniBand mode, what features accelerate AI/HPC workloads on this adapter?

When configured for native InfiniBand NDR mode, this adapter leverages SHARP in-network computing to accelerate collective operations for AI/HPC workloads. It also supports RoCE v2 for low-latency RDMA over 400GbE Ethernet fabrics and GPUDirect RDMA, allowing data to move directly between GPUs and the network card while bypassing the host CPU. Combined with adaptive routing and advanced congestion control, it's exceptionally well suited for the largest-scale AI training clusters, where its single 400Gb/s port maximizes available bandwidth per GPU.

Form Factor	OCP 3.0 Small Form Factor (SFF), tool-less latch
Host Interface	PCIe 5.0 x16
Port Count	1
Port Speed & Connector Type	400Gb/s, QSFP
Network Protocol	VPI - InfiniBand NDR (default mode) or 400GbE Ethernet
RDMA Support	Native InfiniBand RDMA; RoCE v2 in Ethernet mode
Secure Boot	Supported
Crypto (Encryption Offload)	Supported - inline IPsec/TLS hardware offload
Multi-Host	Supported - up to 4 independent hosts share the adapter
Socket Direct	Supported
Controller / ASIC	NVIDIA ConnectX-7
OS Support	Linux (RHEL/Ubuntu/SUSE), Windows Server, VMware ESXi, FreeBSD
Dimensions	Standard/low-profile or OCP 3.0 SFF card, exact dimensions vary by SKU - see datasheet
Operating Temperature	0C to 55C typical
Hardware Warranty	Per NVIDIA/OEM channel warranty terms
Regulatory Certifications	FCC, CE, RoHS (typical for NVIDIA ConnectX adapters)
Management	NVIDIA MFT firmware tools; NC-SI; UEFI HII
Ordering SKU	900-9X7AX-004NMC0

RAID Controllers

HBA（Host Bus Adapters）

Fibre Channel Host HBA

NVMe Adapter

Battery Backup Unit（BBU）

Ethernet Network Adapters

Wireless

Transceiver

Processor／CPU

DRAM／Memory

Server Motherboard

Embedded Motherboard

CPU／AI Accelerator

CPU Expansion Adapter

Video Capture

Stream

Convert

Camera Module

Machine Vision Frame Grabber

Cable

Product Description

Key Features

Technical Specifications

Platform Compatibility

Q&A

Connectivity

Documents & Downloads