Products

Selection and compatibility

Explore NVIDIA Networking switches, adapters, DAC/AOC cables and optical transceivers, plus enterprise wireless products from Aruba, Ruckus, Extreme, Cisco and H3C.

PRODUCT DETAIL

MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment

MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment
MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment
PRODUCT OVERVIEW

MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment

MCX653106A-HDAT Mellanox NVIDIA ConnectX-6 HDR/200GbE IB Network Adapters

  • Product Model MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment
  • Product Categories NVIDIA InfiniBand Adapters
  • Service Support Project delivery and optimization support
View More NVIDIA InfiniBand Adapters Product selection and project support

Product Details

DETAIL MODULES

Deployment answer

MCX653106A-HDAT is a dual-port ConnectX-6 VPI adapter whose retained record lists InfiniBand HDR and Ethernet rates through 200GbE, two QSFP56 ports and PCIe 4.0 x16. It is a candidate when the host design needs two physical network paths or the listed higher-rate modes, but both ports, the host I/O path and the network media must be planned as one system.

Deployment decision dimensions

DimensionRetained recordDeployment check
Complete OPNMCX653106A-HDATQuote, received label and management inventory match
Network portsTwo QSFP56 portsEach port has a defined endpoint, media OPN, mode and failure role
Listed modesHDR through SDR; Ethernet through 200GbESwitch, cable or optic, firmware and FEC support the selected mode
Host interfacePCIe 4.0 x16; 16.0 GT/s is statedSlot width, lane sharing, BIOS and NUMA locality are verified
Adapter familyConnectX-6 VPI with RDMA support statedDriver, RDMA stack, virtualization and monitoring are qualified

Where the dual-port model fits

  • Potential fit: an AI, compute or storage host needs two separately designed HDR or Ethernet paths and has a suitable PCIe 4.0 x16 slot.
  • Test carefully: both ports will carry sustained traffic, cross NUMA boundaries, share a switch failure domain or use mixed media.
  • Not justified by the record alone: the design needs only one link, assumes combined bandwidth without a workload test, or lacks a two-path failure model.

Dual-port versus single-port selection

Compared with the MCX653105A-HDAT page, this record lists two rather than one QSFP56 port and includes HDR and Ethernet rates through 200GbE rather than HDR100 and 100GbE. A second port can support topology choices, but it does not create resilience by itself. Separate switch paths, routing, configuration, power domains and application behavior determine whether the design survives a fault.

Host and network procurement checks

  1. Map both ports to named switch ports, media OPNs, network modes and failure domains.
  2. Verify the server provides the required PCIe 4.0 x16 electrical connection without unexpected lane sharing.
  3. Review CPU and NUMA placement, interrupt policy and workload path for both interfaces.
  4. Approve firmware, driver, operating system, RDMA, virtualization and management versions together.
  5. Keep MCX653106A-HDAT unchanged across quote, bill of materials and acceptance evidence.

Two-port acceptance test

  • Capture device identity, PCIe generation and width, firmware and driver versions.
  • Verify each port independently for mode, rate, FEC where applicable and error counters.
  • Run the representative workload with both intended traffic paths active and observe host and network telemetry.
  • Remove or disable each path in turn, then record failover, recovery and application impact against the declared design.

FAQ

Does a dual-port adapter automatically provide redundancy?

No. Redundancy depends on separate upstream paths, configuration, routing and tested application behavior.

Can both ports be assumed to run at the highest listed rate simultaneously?

No. The retained page lists port and rate capabilities, but the complete host, PCIe, firmware, media, switch and workload combination must be tested.

Should this model replace MCX653105A-HDAT in every new server?

No. Choose by required port count, line rate, slot capacity, validated compatibility and failure design rather than model number alone.

Evidence boundary: This page uses retained model, port, rate, PCIe, family and RDMA values. It does not guarantee aggregate throughput, latency, price, stock, delivery, authorization, certification, warranty, current lifecycle or universal compatibility.

Browse the HDR and Ethernet adapter catalog, compare AI fabric deployment options, examine data-center project contexts, and define the two-port test matrix in the enterprise buyer guide.

Buyer guidance and procurement support

BUYER CHECKLIST
SPEC

Specifications

Confirm model, interface type, rate, power, airflow and software version before ordering.

COMPATIBILITY

Compatibility Boundaries

Technical service and delivery support

PROJECT

Project Conditions

Project delivery and optimization support

TEST

Testing and Validation

Compatibility validation and project risk control

Quick Answers and FAQs

FAQ & NEXT STEP
FIT CHECK

Product selection and project support

Testing and compatibility validation

TEST PATH

Testing and compatibility validation

Project consultation and implementation support

DELIVERY

Solution planning and implementation support

Testing and compatibility validation

FAQ 01

What is MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment suitable for?

MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment can be evaluated for NVIDIA InfiniBand Adapters projects, capacity expansion and replacement planning.

FAQ 02

Buyer guidance and procurement support

Compatibility validation and project risk control

FAQ 03

Testing and compatibility validation

Testing and compatibility validation

FAQ 04

Solution planning and implementation support

Solution planning and implementation support

FAQ 05

How do I confirm the compatibility boundary for MCX653106A-HDAT: Dual-Port HDR and 200GbE Deployment?

Technical service and delivery support

FAQ 06

What information should I provide for a product request?

Model selection and compatibility support