NVIDIA Mellanox® MCA4J80-N003 — Twin-port OSFP 800 Gb/s NDR Active Copper Cable, 3 m
Technical Identity
The NVIDIA Mellanox® MCA4J80-N003 is a 3-metre LinkX active copper cable (ACC) in a twin-port OSFP to twin-port OSFP configuration. Each OSFP head terminates a single Quantum-2-class cage and carries two independent 400 Gb/s NDR InfiniBand ports, so one cable establishes 800 Gb/s aggregate (2 x 400 Gb/s PAM4 lanes) between the two devices it connects. Integrated equalizer / re-timer electronics in each OSFP module extend the practical NDR copper reach from the ~1.5 m passive-DAC ceiling out to 3 m without external power, optics, or fibre. Supplied Refurbished Grade A by T.E.S IT-SOLUTIONS.
Specifications
Manufacturer documentation: NVIDIA MCA4J80-Nxxx 800Gb/s OSFP active copper cable product specifications. T.E.S IT-SOLUTIONS is independent of NVIDIA.
What This Is NOT
NOT a passive DAC. Active electronics in each OSFP head — passive DAC cannot reach 3 m at NDR.
NOT an AOC. Copper conductor; no fibre, no optical transceivers.
NOT a 1-to-2 / 1-to-4 splitter or breakout cable. Each end is a single twin-port OSFP head; 1-to-1 assembly.
NOT a single-port OSFP cable. Both ends are twin-port; will not deliver 800 Gb/s in a single-port-only cage.
NOT an Ethernet cable. Designed and validated for InfiniBand NDR on the NVIDIA Quantum-2 stack.
NOT for legacy HDR / EDR / FDR fabrics. Twin-port OSFP at NDR will not negotiate down to 200 G HDR QSFP56.
NOT a long-reach product. Fixed 3 m length — for runs beyond ~5 m use an AOC or transceiver-plus-fibre.
Compatibility & Constraints
Switch side: NVIDIA Quantum-2 — QM9700 (managed) and QM9790 (externally managed), 32 x twin-port OSFP cages, 1.6 Tb/s per cage. Leaf-spine and switch-to-switch links.
Host side: NVIDIA ConnectX-7 HCAs in twin-port OSFP form factor and NVIDIA BlueField-3 DPUs in NDR / twin-port configurations.
Topology: Intra-rack leaf-to-leaf, short spine-to-leaf in compact AI pods, switch-to-host inside the same row.
Reach: 3 m fixed point-to-point. Longer runs: MFA-prefix AOC or OSFP transceiver pair on MPO/MTP fibre.
Software: MLNX_OFED or NVIDIA DOCA driver stack on hosts; UFM-class fabric management on the IB control plane.
Pre-deployment: Confirm both ends are twin-port-capable cages and that switch firmware advertises NDR PAM4 negotiation on the chosen ports.
Frequently Asked Questions (FAQ)
What is the NVIDIA Mellanox MCA4J80-N003?
The MCA4J80-N003 is an NVIDIA Mellanox LinkX active copper cable (ACC) with twin-port OSFP connectors on both ends, delivering 800 Gb/s aggregate (2 x 400 Gb/s NDR InfiniBand) over a 3-metre copper assembly. It is engineered for NVIDIA Quantum-2 NDR fabrics where the run is too long for a passive DAC but too short to justify an AOC.
How is the MCA4J80-N003 different from a passive DAC at NDR speeds?
At 400 Gb/s NDR per lane the PAM4 signal reach over passive copper is limited to roughly 1.5 metres. The MCA4J80-N003 embeds equalizer / re-timer silicon in each OSFP head so the same copper conductor reliably carries the 800 Gb/s twin-port signal across 3 metres, powered entirely from the host OSFP cages.
Why is this a twin-port OSFP cable, and what does twin-port mean in practice?
Every OSFP cage on the NVIDIA Quantum-2 QM9700 and QM9790 switches is a twin-port cage carrying two independent 400 Gb/s NDR ports. A twin-port OSFP cable mates one cage on each end, so a single MCA4J80-N003 establishes two parallel 400 Gb/s NDR links between the two devices it connects.
Which switches, HCAs and DPUs is the MCA4J80-N003 validated against?
The MCA4J80-N003 targets the NVIDIA NDR ecosystem: Quantum-2 QM9700 and QM9790 InfiniBand switches for leaf-spine and switch-to-switch links, plus host platforms that expose twin-port OSFP cages — typically ConnectX-7 HCAs and BlueField-3 DPUs in twin-port NDR configurations. Verify MLNX_OFED / DOCA versions and port mode on both ends before deployment.
What workload class is this cable right for?
MCA4J80-N003: 800 Gb/s NDR (2 x 400 Gb/s) is built for AI training east-west links, GPU-to-GPU interconnect in DGX/HGX-class clusters, and large-scale HPC fabrics where bisection bandwidth is the binding constraint. It is the wrong tier for ToR/enterprise server access — use 25 GbE / 100 GbE products there.
Where does a 3-metre ACC fit in a typical NDR rack/row plan?
MCA4J80-N003: Three metres is the sweet spot for intra-rack and immediately-adjacent-rack links: leaf-to-leaf within a pod, short spine-to-leaf in compact AI clusters, and switch-to-host runs where a passive DAC will not train. For reach beyond ~5 m step up to an AOC or transceiver-plus-fibre.
Is the MCA4J80-N003 usable in non-NVIDIA OSFP cages?
MCA4J80-N003: Mechanically it complies with the OSFP MSA, but twin-port NDR operation depends on host-side firmware negotiating the 2 x 400 G PAM4 link and is only validated against the NVIDIA Quantum-2 / ConnectX-7 / BlueField-3 stack. For multi-vendor deployments, run a link-quality validation cycle before committing to production.
What condition does T.E.S IT-SOLUTIONS ship?
The MCA4J80-N003 is Refurbished Grade A. Every cable is checked by T.E.S IT-SOLUTIONS before it ships (see Inspection and checks). Ships in anti-static packaging from Haifa, Israel with global B2B logistics.
Why buy from T.E.S IT-SOLUTIONS
T.E.S IT-SOLUTIONS is a Haifa-based NVIDIA Mellanox specialist serving HPC, AI and enterprise data-centre procurement teams since 2011. We focus exclusively on NVIDIA Mellanox, so every cable, adapter and switch is identified by MPN, NVIDIA P/N and ASIC generation before shipping, and every customer gets free pre-sale compatibility advice on Quantum / Spectrum / ConnectX / BlueField pairings. Multi-currency B2B invoicing with worldwide express options, and direct technical contact with our engineering team. See tes-itsolutions.com.