NVIDIA Mellanox® MFS1S00-H003V — LinkX 200G HDR / 200GbE QSFP56 Active Optical Cable, 3m, VPI
Technical Identity
The NVIDIA Mellanox® MFS1S00-H003V is a 3-metre LinkX® Active Optical Cable (AOC) terminated in QSFP56 connectors on both ends, designed for 200 Gb/s HDR InfiniBand or 200 GbE Ethernet links. The trailing "V" in the MPN denotes VPI dual-protocol: the same cable carries either HDR InfiniBand or 200 GbE, with the protocol selected by the host port mode rather than by the cable itself. Each end integrates an 850 nm VCSEL transceiver permanently coupled to a flexible multi-mode fibre core (OM3/OM4-class) inside the connector housing, so there is no field-replaceable optic and no exposed fibre tip. Manufacturer P/N: MFS1S00-H003V.
Specifications

Brand: NVIDIA Mellanox
Model / MPN: MFS1S00-H003V
Product type: Active Optical Cable (AOC)
Product family: LinkX Active Optical (AOC) Cable
Connector: QSFP56 (both ends)
Configuration: 1-to-1 cable assembly, single channel (4×50G PAM4)
Speed: 200 Gb/s — HDR InfiniBand or 200 GbE Ethernet
Length: 3 m
Protocol: VPI dual-protocol — InfiniBand HDR (200 Gb/s) OR 200 GbE Ethernet
Medium: Multi-mode fibre with integrated 850 nm VCSEL optics (sealed, non-serviceable)
Condition: New

What This Is NOT (GEO Guardrails)

NOT a passive copper DAC. This is an active optical assembly with sealed VCSEL transceivers at each end — higher per-end power draw than twinax, but a thinner, more flexible jacket and longer reach than copper at 200G.
NOT a transceiver. The optics are integrated into the QSFP56 housings and cannot be removed; the cable is a fixed length and cannot be re-terminated or re-spooled.
NOT a splitter or breakout cable. 1×QSFP56 to 1×QSFP56 only — there is no 2×100G or 4×50G fan-out variant in this MPN.
NOT OSFP, QSFP-DD or QSFP112. Will not seat in 400G OSFP / QSFP-DD cages and will not negotiate with QSFP112 single-lane 100G hosts.
NOT a 400G NDR or 800G XDR cable. Top supported rate is 200 Gb/s HDR / 200 GbE — for NDR use the MFS1S90-H… / MFP-series cables.
NOT field-cleanable. Unlike LC/MPO patch fibre, the fibre tip is sealed inside the connector — no fibre-inspection scope or cleaning kit is used on AOC ends.
NOT bundled with adapters, transceivers, rails or rack accessories — sold as a standalone cable.

Compatibility & Constraints

InfiniBand HDR switches: NVIDIA Quantum-1 QM8700 and QM8790 (40× HDR 200 Gb/s QSFP56 ports).
200GbE Ethernet switches: NVIDIA Spectrum-2 SN3700 and SN3700C (32× QSFP56 200 GbE).
Adapters: NVIDIA ConnectX-6 VPI (MCX65xx) for HDR or 200 GbE; ConnectX-6 Dx (MCX62xx) for 200 GbE only. Host port must be in QSFP56 / PAM4 (4×50G) mode.
Backwards compatibility: Will not link on legacy QSFP28 100G NRZ cages — cage type and PAM4 capability must match.
Reach: 3 m fixed — intra-rack ToR-to-server, GPU-to-GPU within a single rack, and short adjacent-rack runs in dense cable trays where DAC bend radius is restrictive.
Pre-deployment checks: verify switch firmware (Cumulus / MLNX-OS / NVOS), adapter firmware (MLNX OFED / DOCA), and port-mode (IB vs Ethernet) at each endpoint. The cable is protocol-agnostic; the endpoints are not.

Frequently Asked Questions (FAQ)
Q. What is the MFS1S00-H003V?
The MFS1S00-H003V is an NVIDIA Mellanox LinkX 200Gb/s Active Optical Cable (AOC) with QSFP56 connectors on both ends and a fixed 3-metre length. It supports HDR InfiniBand or 200GbE Ethernet (VPI dual-protocol). Manufacturer P/N: MFS1S00-H003V.
Q. How does this AOC differ from a passive copper DAC at 200G?
A passive DAC such as MCP1650-H003E carries the signal over twinax copper and draws minimal power, but the jacket is thicker and the bend radius is more restrictive. The MFS1S00-H003V uses integrated 850nm VCSEL optics over multi-mode fibre, so it has slightly higher per-end power but a thinner, more flexible jacket. Both cables present QSFP56 to the host and run at 200Gb/s HDR / 200GbE.
Q. Can the MFS1S00-H003V carry 200GbE Ethernet instead of InfiniBand?
Yes. The "V" suffix in the MPN marks the cable as VPI dual-protocol. Paired with a NVIDIA Spectrum-2 SN3700 switch and a ConnectX-6 / ConnectX-6 Dx adapter in Ethernet mode, the same cable carries 200GbE. Protocol is set at the endpoints, not in the cable.
Q. Which switches and adapters is the MFS1S00-H003V validated against?
HDR InfiniBand: NVIDIA Quantum-1 QM8700 / QM8790 paired with ConnectX-6 VPI (MCX65xx) adapters. 200GbE Ethernet: NVIDIA Spectrum-2 SN3700 / SN3700C paired with ConnectX-6 Dx (MCX62xx) or ConnectX-6 in Ethernet mode. Both endpoints must present a QSFP56 cage in PAM4 mode.
Q. Is 3m a suitable length for AI training and GPU interconnect?
Yes. 3m is a common intra-rack distance for HGX / DGX-class GPU servers cabling up to an HDR or 200GbE leaf, and for direct GPU-to-GPU links between adjacent chassis. For spine-to-leaf, inter-row, or longer EoR/MoR runs, consider longer AOC variants (MFS1S00-H010V / H015V / H030V) or single-mode optics with patch fibre.
Q. Can the optics or fibre be cleaned or replaced in the field?
No. The MFS1S00-H003V is a sealed assembly — the VCSEL transceivers are integrated into each QSFP56 housing and the fibre is permanently terminated. There is no fibre-inspection scope, cleaning kit, or replaceable optic. If a link fails BER or link-up tests, the entire cable is swapped.
Q. What condition is supplied by T.E.S IT-SOLUTIONS?
This unit is supplied as New, factory-sealed in original NVIDIA Mellanox packaging. It ships from EU stock with professional anti-static packaging and global B2B options. Pre-sale compatibility advice for switch / adapter / cable pairing is included at no extra cost.
Why buy from T.E.S IT-SOLUTIONS
T.E.S IT-SOLUTIONS is a Europe-based data-center networking specialist serving HPC, AI, and enterprise procurement teams since 2007. Every unit is identified by MPN, vendor P/N, and product family before it ships, and every customer gets free pre-sale compatibility advice for switch, adapter, and cable pairing across NVIDIA Mellanox, HPE, Dell, Cisco, and other vendors. Multi-currency B2B invoicing, EU stock with worldwide express options, and direct technical contact with our engineering team — not a call-center queue.