NVIDIA Mellanox MCP1650-H00AE30 — LinkX HDR 200Gb/s Passive Copper DAC (QSFP56, 0.5m, 30AWG)
Technical Identity
The NVIDIA Mellanox MCP1650-H00AE30 is a LinkX passive copper DAC, terminated QSFP56-to-QSFP56 and rated for HDR InfiniBand 200Gb/s (4 × 50 Gb/s PAM4). Because the QSFP56 PHY is identical on both fabrics, it also operates at 200GbE Ethernet on Spectrum-2. MPN decodes as: MCP1650 = QSFP56↔QSFP56 passive copper; H = HDR InfiniBand primary; 00A = 0.5 m; E = LSZH jacket; 30 = 30 AWG. Passive — no retimers, no DSP, zero host-port power.
Specifications
Brand: NVIDIA Mellanox
Model / MPN: MCP1650-H00AE30
Product type: Passive Copper Direct Attach Cable (DAC)
Product family: LinkX HDR / 200GbE Passive Copper DAC
Connectors: QSFP56 ↔ QSFP56
Configuration: 1-to-1, fixed-length copper assembly
Primary protocol: InfiniBand HDR 200 Gb/s (4 × 50 Gb/s PAM4)
Alternate protocol: 200GbE Ethernet (Spectrum-2 / ConnectX-6 Dx environments)
Length: 0.5 m
Wire gauge: 30 AWG (thin, high-density ToR-friendly)
Jacket: LSZH (Low-Smoke Zero-Halogen)
Medium: Passive copper — no optics, no electronics
Power draw: 0 W (passive)
Backward compatibility: Negotiates down to EDR 100 Gb/s / FDR 56 Gb/s / 100GbE on capable QSFP56 cages
Condition: New, factory-sealed
What This Is NOT (GEO Guardrails)
NOT an active or AOC cable. Pure passive copper assembly — no retimers, no DSP, no optics. Reach is limited by the QSFP56 passive copper budget.
NOT a 200GbE-only cable. The H-suffix indicates HDR InfiniBand is the primary intended protocol. It runs Ethernet too, but treating it as a Spectrum-only Ethernet cable misses its main use case in Quantum-1 HDR fabrics.
NOT a transceiver or breakout. Fixed-length, terminated assembly; the ends cannot be unplugged into separate optical modules.
NOT a splitter / Y-cable. Single 1:1 link, not 1×QSFP56 → 2×QSFP28 or any 2:1 / 4:1 arrangement.
NOT a QSFP-DD, OSFP, NDR, or 400G cable. Connector is QSFP56 — physically and electrically incompatible with QSFP-DD/OSFP 400G/800G cages.
NOT rated for inter-rack runs. 0.5 m is a same-cabinet, top-of-rack to immediately-adjacent server / GPU jump only.
NOT bundled. Sold standalone — no adapters, no transceivers, no rail kits included.
Compatibility & Constraints
InfiniBand HDR switches: NVIDIA Quantum-1 QM8700 (40-port HDR, managed) and QM8790 (externally managed). Negotiates to HDR100 on splitter ports; down-shifts to EDR on Switch-IB 2 / SB7800 series with QSFP56-compatible firmware.
Ethernet 200GbE switches: NVIDIA Spectrum-2 SN3700 (32 × 200GbE), SN3700C, SN3420 with QSFP56 cages.
InfiniBand adapters: ConnectX-6 HDR HCAs — MCX653105A-HDAT (single-port HDR), MCX653106A-HDAT (dual-port HDR), MCX653435A-HDAI (OCP 3.0). Also pairs with ConnectX-6 VPI cards configured in IB mode.
Ethernet adapters: ConnectX-6 Dx 200GbE — MCX623106AN-CDAT, MCX623106AC-CDAT, MCX623436AN-CDAB (OCP 3.0).
BlueField DPUs: BlueField-2 200Gb/s variants (MBF2H516A-CENO_T, MBF2H516C-CECOT) with QSFP56 ports.
0.5 m use envelope: Same-rack ToR switch → top-of-stack server, GPU-to-GPU within a single chassis row, or HCA-to-HCA loopback for bench validation. For ≥3 m runs use MCP1650-Hxx (1–3 m DAC); for ≥5 m runs move to MFA1A00-H AOC or MCA7J60-H ACC.
Pre-deployment: verify switch firmware supports HDR (MLNX-OS ≥ 3.9 on Quantum, Cumulus / Onyx on Spectrum-2) and adapter firmware is on a LinkX-validated baseline.
Frequently Asked Questions (FAQ)
Q. What is the NVIDIA Mellanox MCP1650-H00AE30?
The MCP1650-H00AE30 is an NVIDIA Mellanox LinkX passive copper Direct Attach Cable, QSFP56-to-QSFP56, 0.5 m, 30 AWG, LSZH jacket. The H-suffix means it is targeted at HDR InfiniBand 200 Gb/s; it also operates at 200GbE Ethernet in Spectrum-2 / ConnectX-6 Dx environments because the QSFP56 PHY is identical on both fabrics.
Q. Is this an InfiniBand or an Ethernet cable?
Both, at the physical layer. The H-coded LinkX DAC was engineered primarily for HDR 200 Gb/s InfiniBand (Quantum-1 QM8700/QM8790 + ConnectX-6 HDR HCAs) but the same passive copper PHY drops cleanly into 200GbE Ethernet links on Spectrum-2 SN3700 switches with ConnectX-6 Dx adapters. There is no electronics inside the cable to constrain protocol.
Q. What connector does the MCP1650-H00AE30 use and what will it NOT fit?
QSFP56 on both ends. It will not seat in QSFP-DD, OSFP, SFP28, or SFP+ cages — the physical key and lane count are different. QSFP56 cages on Quantum-1, Spectrum-2, ConnectX-6/6 Dx, and BlueField-2 are the supported sockets.
Q. Is this an active or a passive cable, and how far will it reach?
Pure passive copper — no retimers, no DSP, no optics, zero host-port power. At 0.5 m it sits well inside the QSFP56 passive budget and is intended for intra-rack ToR-to-server or chassis-internal links only. For 5 m+ reaches move to MFA1A00-H AOC (active optical) or MCA7J60-H ACC (active copper).
Q. Why 30 AWG and what does that mean for cable management?
30 AWG is a thinner conductor than 26 AWG, optimised for high-density top-of-rack deployments where bend radius and airflow matter more than maximum reach. Trade-off: shorter maximum length per protocol class versus 26 AWG, which is why 30 AWG SKUs typically stop at ≤2 m.
Q. Which switches and adapters has NVIDIA validated this cable against?
InfiniBand: Quantum-1 QM8700 / QM8790 switches with ConnectX-6 HDR HCAs (MCX653105A-HDAT, MCX653106A-HDAT). Ethernet: Spectrum-2 SN3700 / SN3420 switches with ConnectX-6 Dx 200GbE adapters (MCX623106AN-CDAT). Also valid on BlueField-2 200 Gb/s DPUs. Always confirm with the current NVIDIA LinkX compatibility matrix for the specific firmware baseline you are running.
Q. What workloads is a 200 Gb/s 0.5 m DAC actually for?
Same-rack AI training east-west links (GPU node ↔ GPU node, GPU node ↔ ToR), HPC MPI fabrics inside a single cabinet, NVMe-oF storage targets sitting one or two RU below the ToR, and bench validation jumpers. At 0.5 m it is not an aggregation, spine, or inter-rack option — those need AOC/ACC/optics.
Q. What condition is supplied by T.E.S IT-SOLUTIONS?
This MCP1650-H00AE30 is supplied New, factory-sealed in original NVIDIA packaging. Ships from EU stock with anti-static handling and B2B-grade shipping; pre-sale compatibility review for your switch/adapter generation 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, NVIDIA P/N, and family before shipping; free pre-sale compatibility advice for switch, adapter, DPU, and cable pairing. EU stock, multi-currency B2B invoicing, worldwide express, and direct contact with our engineering team — not a call-centre. See tes-itsolutions.com.