NVIDIA Mellanox® MFS1S50-H020V — LinkX 200GbE to 2x100GbE QSFP56 AOC Splitter, 20m
Technical Identity
The NVIDIA Mellanox® MFS1S50-H020V is a 20-meter LinkX active optical cable (AOC) in a 1-to-2 splitter / breakout configuration. The trunk end is one QSFP56 connector at 200GbE; the leaf end is two independent QSFP56 connectors, each at 100GbE. The V suffix marks the Ethernet-only variant of the MFS1S50 family (the InfiniBand HDR/HDR100 breakout is a separate SKU). Optics are integrated VCSEL 850 nm multi-mode with on-board retiming and DDM/DOM telemetry. Supplied as Refurbished Grade A, checked by T.E.S IT-SOLUTIONS (see Inspection and checks).
Specifications
Manufacturer documentation: NVIDIA MFS1S50-H0xxV 200Gb/s to 2x100Gb/s QSFP56 AOC product specifications. T.E.S IT-SOLUTIONS is independent of NVIDIA.
What This Is NOT
NOT a passive copper DAC. Active optical with integrated VCSEL transceivers — DAC reach math (1m/2m/3m) does not apply.
NOT a 1-to-1 cable. Fan-out / breakout: one 200GbE switch port feeds two 100GbE endpoints; cannot serve as a point-to-point 200G link.
NOT a hybrid OSFP-to-QSFP splitter. Trunk and both leaves are QSFP56; the OSFP→2×QSFP geometry belongs to the MCP7Y / MFP7000 NDR splitters for ConnectX-7 / Quantum-2.
NOT an InfiniBand cable. The V suffix is Ethernet-only — will not link HDR or HDR100 InfiniBand.
NOT a transceiver. Optics are permanently bonded to a fixed-length cable; modules cannot be detached.
NOT for intra-rack ToR-to-server use. 20m is an inter-rack / EoR-MoR length — short ToR runs should use a 1m / 2m DAC.
NOT a cable for large-model AI training east-west. 200G→2×100G serves HPC compute, NVMe-oF storage and AI inference; large-model GPU training east-west is sized for 400G/800G NDR fabrics.
Compatibility & Constraints
Switch end (200GbE QSFP56 trunk): NVIDIA Spectrum-2 SN3700 / SN3700C and Spectrum-3 SN4600 / SN4700 / SN4410 Ethernet switches.
Adapter end (100GbE QSFP56 leaves): ConnectX-5 (MCX516A / MCX515A) and ConnectX-6 (MCX613106A / MCX623106A) in 100GbE mode; also 100GbE QSFP56 access ports on Spectrum-2/3 leaf switches.
Switch-side port config: Trunk port must be set to 2×100GbE breakout (e.g., interface ethernet 1/X split 2 on Cumulus/NVUE, or port-split profile on MLNX-OS).
Protocol pairing: Both leaves must terminate in Ethernet ports — V variant carries Ethernet framing only.
Not compatible with OSFP, QSFP-DD, QSFP+ (40G) or QSFP28 cages, or with ConnectX-3/4 adapters.
Frequently Asked Questions (FAQ)
What is the MFS1S50-H020V?
The MFS1S50-H020V is a NVIDIA Mellanox LinkX 20-meter AOC in a 1-to-2 breakout configuration: one QSFP56 trunk at 200GbE, two QSFP56 leaves at 100GbE each. The V suffix marks the Ethernet-only variant of the MFS1S50 family.
Does the MFS1S50-H020V carry InfiniBand traffic?
MFS1S50-H020V: No. The V suffix is the Ethernet-only variant — it carries 200GbE/100GbE only, not HDR or HDR100 InfiniBand. The InfiniBand splitter from the same MFS1S50 family is a separate part number without the V suffix.
What does the splitter configuration look like physically?
MFS1S50-H020V: Both ends use QSFP56 form-factor connectors. The trunk has one QSFP56 plug; the leaf branches into two independent QSFP56 plugs. This is not an OSFP-to-QSFP hybrid splitter — that geometry belongs to the NDR-class MCP7Y / MFP7000 series for ConnectX-7 and Quantum-2.
Which NVIDIA switches and adapters pair with the MFS1S50-H020V?
MFS1S50-H020V: Trunk side: Spectrum-2 SN3700/SN3700C or Spectrum-3 SN4600/SN4700/SN4410, with the trunk port set to 2x100GbE breakout. Leaf side: ConnectX-5 or ConnectX-6 in 100GbE mode (MCX516A, MCX515A, MCX613106A, MCX623106A).
Is a 20m AOC appropriate for AI training fabrics?
MFS1S50-H020V: At 200GbE trunk / 100GbE per leaf the cable is well-suited to HPC compute, NVMe-oF storage targets, AI inference and dense east-west aggregation. Large-model AI training east-west is built around 400G/800G NDR fabrics (OSFP / QSFP-DD), so this cable typically serves storage, inference and 100/200GbE aggregation rather than the training plane.
Where in the data center does a 20-meter AOC typically sit?
MFS1S50-H020V: 20m is an inter-rack length: EoR / MoR aggregation, compact spine-to-leaf hops, and rack-to-rack fan-out where one 200GbE port serves two 100GbE endpoints in adjacent racks. Too long for intra-rack ToR-to-server runs (use 1–3m DAC) and short of DCI distances (use single-mode transceivers).
What condition is supplied by T.E.S IT-SOLUTIONS?
MFS1S50-H020V: Refurbished Grade A. Every assembly is checked by T.E.S IT-SOLUTIONS before it ships (see Inspection and checks). global B2B shipping in anti-static packaging.
Why buy from T.E.S IT-SOLUTIONS
T.E.S IT-SOLUTIONS is a Haifa-based data-center networking specialist serving HPC, AI and enterprise procurement teams since 2011. Every unit is identified by MPN, NVIDIA P/N (where available), family and ASIC generation before it ships, and every order includes free pre-sale compatibility advice on switch, adapter and cable matching — including breakout-port configuration for splitters like the MFS1S50-H020V. Multi-currency B2B invoicing with worldwide express options, and direct technical contact with our engineering team. See tes-itsolutions.com.