NVIDIA Mellanox MTQ8200-HS2F Quantum MetroX-2 Long-Haul InfiniBand Switch - 2 x 100Gb/s EDR long-haul ports to 40km plus 8 x 200Gb/s HDR local ports, 1U, managed, P2C airflow. The MTQ8200-HS2F (platform TQ8200-HS2F) is a purpose-built long-haul InfiniBand switch that carries native InfiniBand between physically separate data centres. NVIDIA describes the OPN as a MetroX-2 40km long-haul 100Gb/s switch with 2 long-haul QSFP28 ports, 8 standard HDR ports, 2 AC power supplies, x86 dual-core management, standard depth, P2C airflow and rail kit. It is a site-to-site extension switch, not a general-purpose top-of-rack switch. Refurbished, supplied by T.E.S IT-SOLUTIONS.
Engineering Context
MetroX-2 is built on the NVIDIA Quantum HDR 200Gb/s InfiniBand switch device, so it extends native InfiniBand semantics - RDMA, adaptive routing, congestion control and SHARP collective offload - across metro distance without gateways or encapsulation. The two long-haul ports are QSFP28 running EDR 100Gb/s, graded to 40km on the TQ8200 (the TQ8100 sibling is graded to 10km); the eight local ports are QSFP56 running HDR 200Gb/s. NVIDIA lists a maximum system throughput of 4Tb/s across the 2 x QSFP28 and 8 x QSFP56 non-blocking ports. The platform is IBTA 1.21 and 1.3 compliant with 256-byte to 4Kbyte MTU, port mirroring, VL2VL mapping and SHIELD self-healing. Distance dominates latency: light in single-mode fibre propagates at roughly 5ns per metre, so a 40km span adds approximately 200 microseconds of one-way delay regardless of switch performance. Management is an x86 COMEX Broadwell dual-core CPU running MLNX-OS, with a web GUI giving alarms, event history and performance monitoring for every optical module.
Deployment and Use Cases
Multi-site HPC clusters where compute and storage sit in different buildings or campuses. Business continuity and disaster recovery designs requiring zero Recovery Point Objective with native RDMA between sites. Replication and mirroring for Lustre, IBM Spectrum Scale (GPFS) and BeeGFS across metro distance. Migrating batch jobs between InfiniBand data centres, or federating two remote clusters into one fabric. Campus connectivity between research buildings on dedicated dark fibre.
Technical Specifications
Brand / MPN: NVIDIA Mellanox, MTQ8200-HS2F (platform TQ8200-HS2F)
Family: NVIDIA Quantum MetroX-2 long-haul systems
Switch device: NVIDIA Quantum HDR 200Gb/s InfiniBand ASIC
Long-haul ports: 2 x QSFP28, EDR 100Gb/s, reach up to 40km
Local ports: 8 x QSFP56, HDR 200Gb/s
Max system throughput: 4Tb/s (per NVIDIA TQ8x00 user manual)
Protocol: InfiniBand only (EDR long-haul, EDR/HDR local)
Features: RDMA over distance, SHARP, adaptive routing, congestion control, SHIELD, port mirroring, VL2VL mapping
Management: internally managed - x86 COMEX Broadwell dual-core, MLNX-OS, CLI, Web UI, SNMP v1/v2/v3
Subnet manager: on-board SM supporting fabrics of up to 2,000 nodes; UFM agent
Management ports: 100/1000Mb/s RJ45 Ethernet, RS232 console, USB
Airflow: P2C (power-to-connector) forward - air enters at the PSU side, exits at the port side
Power supplies: 2 x AC, dual redundant hot-plug slots, 1+1; 100-127VAC / 200-240VAC, 50-60Hz
Fans: hot-swappable, N+1 redundant
Form factor: 1U 19in rack-mountable, standard depth, rail kit included
Environment: 0 C to 40 C operating, up to 3,200m, RoHS compliant
Condition: Refurbished - tested by T.E.S IT-SOLUTIONS
Compatibility, Optics and Parent Systems
The long-haul ports require NVIDIA LR4 MetroX-2 QSFP28 transceivers and a dedicated single-mode dark-fibre pair per link, ordered separately and matched to the actual span; 40km is the maximum rating and real reach depends on fibre loss budget, splices and patch panels. The eight local HDR QSFP56 ports attach to NVIDIA Quantum HDR fabrics and to ConnectX-6 and ConnectX-7 adapters at 200Gb/s using QSFP56 DACs (MCP1650-H), AOCs (MFS1S00-H) or HDR transceivers; ConnectX-5 endpoints connect at EDR 100Gb/s. Typical local head-end switches are the 40-port MQM8700 and MQM8790. Sibling: MTQ8100-HS2F, the 10km version. Predecessor generation: MetroX TX6100 / TX6240 / TX6280 FDR10-class long-haul. Confirm SM ownership, firmware and MLNX-OS versions across both sites before cutover.
What This Is NOT
NOT a general-purpose top-of-rack switch. Only 8 local ports. For dense HDR aggregation use MQM8700 / MQM8790 (40 ports).
NOT an Ethernet, DWDM or IP-over-WAN device. It carries native InfiniBand over dark fibre and does not bridge to Ethernet.
NOT 200Gb/s over distance. The long-haul ports are EDR 100Gb/s QSFP28; only the 8 local ports run HDR 200Gb/s.
NOT the 10km TQ8100. MTQ8200-HS2F is the 40km model; reach grading is not field-changeable.
NOT a C2P airflow unit. The F suffix is P2C forward airflow - air enters at the PSU side and exits at the port side.
NOT NDR-capable, and NOT supplied with long-haul optics, fibre or software entitlements unless listed on the quotation.
Frequently Asked Questions
Q. How far can the MTQ8200-HS2F extend an InfiniBand fabric?
Up to 40km. The two QSFP28 long-haul ports run EDR 100Gb/s and are graded to 40 kilometres over single-mode fibre. The 10km version of the same platform is the MTQ8100-HS2F. Native RDMA, adaptive routing and SHARP are preserved across the span.
Q. What optics and fibre does the MTQ8200-HS2F need?
NVIDIA LR4 MetroX-2 QSFP28 transceivers and a dedicated single-mode dark-fibre pair per long-haul link, sourced separately. Reach must be validated against the real loss budget of the span including splices and patch panels; 40km is the maximum rating, not a guarantee for every fibre plant.
Q. How many ports does the MTQ8200-HS2F have and at what speeds?
Ten in total: 2 x QSFP28 long-haul ports at EDR 100Gb/s and 8 x QSFP56 local ports at HDR 200Gb/s. NVIDIA lists a maximum system throughput of 4Tb/s for the TQ8x00 platform.
Q. Is the MTQ8200-HS2F managed, and can it run the subnet manager?
Yes. It is internally managed with an x86 COMEX Broadwell dual-core CPU running MLNX-OS, and carries an on-board subnet manager supporting fabrics of up to 2,000 nodes, plus CLI, Web UI and SNMP. It can also be operated under NVIDIA UFM. Decide which site owns the SM before cutover.
Q. Is the MTQ8200-HS2F port-to-connector or connector-to-power airflow?
P2C - power-to-connector, forward airflow. Air enters at the power-supply side and exhausts at the port side. PSU and fan modules are airflow-specific and must all match the chassis direction.
Why T.E.S IT-SOLUTIONS
T.E.S IT-SOLUTIONS (tes-itsolutions.com) supplies refurbished NVIDIA Mellanox InfiniBand hardware to B2B, HPC and research customers across the EU and worldwide. Every MetroX unit is bench-verified before shipment: power-on, management response, port and optical-module status, PSU and fan redundancy check and cosmetic inspection. EU stock, 1-3 business day handling and technical consultation on long-haul InfiniBand topology and optics selection.