NVIDIA Mellanox MTX6100-2SFS MetroX TX6100 10km Long-Haul FDR10 InfiniBand Switch, 1U, managed, P2C airflow. The MTX6100-2SFS is a 1U long-haul InfiniBand switch from the Mellanox MetroX family, built to carry native InfiniBand over dedicated dark fibre between sites up to 10 kilometres apart. Long-haul ports run FDR10 at 40Gb/s per port over QSFP+ cages fitted with long-reach LR4-type transceivers. The vendor ordering string reads: MetroX 10km long-haul 40G switch, 6 QSFP ports, 2 AC power supplies, PPC460 management processor, standard depth, P2C airflow, rail kit, RoHS-6. This is a site-to-site fabric extension switch, not a top-of-rack cluster switch. Refurbished, supplied by T.E.S IT-SOLUTIONS.
Engineering Context
MetroX carries InfiniBand semantics - RDMA, low-latency send/receive and subnet routing - across metro distance without protocol conversion or WAN encapsulation. FDR10 runs four lanes per QSFP+ port at 10.3125Gb/s with 64/66b encoding for 40Gb/s of usable bandwidth, which is why FDR10 delivers more payload than QDR from the same 40Gb/s signalling rate (QDR uses 8/10b encoding and yields 32Gb/s). Mellanox launch documentation for the TX6100 describes six long-haul ports at 40Gb/s giving 240Gb/s of aggregate long-haul throughput, together with six data-centre downlink QSFP ports running at 40 or 56Gb/s for 336Gb/s aggregate; the NVIDIA ordering string quotes 6 QSFP ports, so confirm the exact faceplate population of the unit being quoted. Mellanox quoted a port-to-port hop latency of approximately 200ns inside the switch, with roughly 5ns per metre added by the fibre itself - about 50 microseconds of one-way propagation delay on a full 10km span. Management runs on a PowerPC PPC460 processor with CLI and SNMP.
Deployment and Use Cases
Campus and metro federation of two InfiniBand clusters into a single fabric. Storage replication and mirroring for Lustre, IBM Spectrum Scale (GPFS) and BeeGFS between buildings. Disaster recovery and business continuity topologies that must preserve RDMA rather than fall back to TCP. Research backbones linking laboratory and machine-room sites on university dark fibre. Low-latency cross-site links in financial and simulation environments where a 10km hop is the physical constraint.
Technical Specifications
Brand / MPN: NVIDIA Mellanox, MTX6100-2SFS
Family: Mellanox MetroX TX6100 long-haul systems
Product type: 1U managed long-haul InfiniBand switch
Long-haul rate: FDR10, 40Gb/s per port (4 x 10.3125Gb/s lanes, 64/66b encoding)
Reach: up to 10km over single-mode dark fibre
Connector: QSFP+ (long-reach LR4-type transceivers required, not supplied)
Long-haul aggregate: 240Gb/s across six long-haul ports (per Mellanox launch documentation)
Local downlink rate: 40Gb/s FDR10 or 56Gb/s FDR on QSFP+ (336Gb/s aggregate)
Switch hop latency: approximately 200ns, plus approximately 5ns per metre of fibre
Protocol: InfiniBand only
Management: managed - PowerPC PPC460 processor, CLI, SNMP
Power supplies: 2 x AC, redundant
Airflow: P2C (power-to-connector) - air enters at the PSU side and exits at the port side
Form factor: 1U rack-mountable, standard depth, rail kit
Compliance: RoHS-6
Condition: Refurbished - tested by T.E.S IT-SOLUTIONS
Compatibility and Parent Systems
The MTX6100-2SFS belongs to the FDR/FDR10 generation of Mellanox InfiniBand. Typical endpoint adapters are ConnectX-2 (QDR) and ConnectX-3 / ConnectX-3 Pro (FDR10 and FDR 56Gb/s) using QSFP+ cabling; later ConnectX cards negotiate down where the port profile permits. Typical local head-end switches of the same era are the SX6036 and SX6025 36-port FDR switches and the SX6512 / SX6536 directors. Long-haul ports need matched long-reach QSFP+ optics and a dedicated single-mode fibre pair per link. Family siblings by reach: TX6100 (10km), TX6240 (40km), TX6280 (80km). Airflow sibling: MTX6100-2SRS is the reverse-airflow (C2P) variant. Successor generation: NVIDIA Quantum MetroX-2, MTQ8100-HS2F (10km) and MTQ8200-HS2F (40km), which raise the local ports to HDR 200Gb/s and the long-haul ports to EDR 100Gb/s. Verify firmware, subnet manager ownership and optic compatibility across both sites before cutover.
What This Is NOT
NOT an FDR, EDR, HDR or NDR long-haul switch. The long-haul rate is FDR10 40Gb/s. EDR 100Gb/s long-haul requires MetroX-2 (MTQ8100 / MTQ8200).
NOT a general-purpose top-of-rack switch. The radix is designed for long-haul aggregation, not for dense cluster switching.
NOT an Ethernet, DWDM or IP-over-WAN device. It carries native InfiniBand over dark fibre and does not bridge to Ethernet.
NOT a QSFP28 or QSFP56 platform. Ports are QSFP+; QSFP28 and QSFP56 modules will negotiate down at best and are not the intended media.
NOT a C2P airflow unit. The F in -2SFS is P2C forward airflow; the reverse-airflow model is MTX6100-2SRS. PSU and fan modules must share one airflow direction.
NOT supplied with long-reach optics, fibre, patch leads or software entitlements unless listed on the quotation.
Frequently Asked Questions
Q. How far can the MTX6100-2SFS extend an InfiniBand fabric?
Up to 10 kilometres over single-mode dark fibre, at FDR10 40Gb/s per long-haul port. The 40km and 80km members of the same MetroX family are the TX6240 and TX6280.
Q. What optics and fibre does the MTX6100-2SFS require?
Long-reach LR4-type transceivers in the QSFP+ cages plus a dedicated single-mode dark-fibre pair per long-haul link, sourced separately. The 10km figure is a maximum rating; validate it against the real loss budget of the span, including splices and patch panels.
Q. Is the MTX6100-2SFS managed?
Yes. The MTX6100-2SFS is a managed switch running on a PowerPC PPC460 management processor with CLI and SNMP access for configuration, monitoring and firmware management. Fabric-wide subnet management still has to be assigned explicitly across the two sites.
Q. Is the MTX6100-2SFS Power-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. The reverse-airflow model is the MTX6100-2SRS. All PSU and fan modules in the chassis must share the same airflow direction.
Q. What latency does a 10km MetroX link add?
The switch hop itself is approximately 200ns. Fibre propagation dominates at roughly 5ns per metre, so a full 10km span adds about 50 microseconds of one-way delay. Size replication and checkpoint windows around the fibre delay, not the switch.
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 long-haul unit is bench-verified before shipment: power-on, management response, port and PSU status and full cosmetic inspection. EU stock, 1-3 business day handling and technical consultation on long-haul InfiniBand topology, optics selection and fibre budget at no extra cost.