NVIDIA Mellanox® 900-9X81Q-00SN-ST0 — ConnectX-8 SuperNIC, 400GbE / XDR400, Single-Port QSFP112, PCIe Gen6 x16

Technical Identity
The NVIDIA Mellanox® 900-9X81Q-00SN-ST0 is a ConnectX-8 SuperNIC built on the ConnectX-8 ASIC in a Standard PCIe HHHL (Half-Height, Half-Length) form factor. It is a single-port QSFP112 adapter delivering up to 400GbE (Ethernet) or XDR400 — NDR 400 Gb/s InfiniBand (VPI dual-protocol, selectable per port mode). The host interface is PCIe Gen6 x16, fully backward-compatible with PCIe Gen5 / Gen4 x16 slots. Condition: Refurbished Grade A — bench-tested by T.E.S IT-SOLUTIONS.

Specifications

OEM: NVIDIA Mellanox
Part Number (MPN): 900-9X81Q-00SN-ST0
Product type: ConnectX-8 SuperNIC (PCIe adapter card)
ASIC / Family: ConnectX-8
Port configuration: Single-port QSFP112
Ethernet speeds: 400 / 200 / 100 / 50 / 40 / 25 / 10 GbE
InfiniBand speeds: XDR400 (NDR 400 Gb/s) / NDR200 / HDR / HDR100 / EDR
Protocol mode: VPI dual-protocol — Ethernet or InfiniBand selectable
Host interface: PCIe Gen6 x16 (backward compatible with Gen5 / Gen4 x16)
Form factor: Standard PCIe HHHL, tall bracket (low-profile not included)
Condition: Refurbished Grade A — bench-tested by T.E.S IT-SOLUTIONS


What This Is NOT

NOT an OSFP variant. This SKU is QSFP112 only. The OSFP-cage SuperNIC variant is 900-9X81Q-00CN-ST0 (800 Gb/s, single-port OSFP) — do not confuse cages.
NOT a dual-port card. 9X81Q-00SN is single-port; dual-port variants use a different MPN suffix.
NOT a DPU / BlueField. ConnectX-8 SuperNIC is a fixed-function RDMA NIC — it does not host Arm cores, Linux, or DOCA platform services like BlueField-3 does.
NOT a ConnectX-7 or earlier generation. ConnectX-8 introduces PCIe Gen6 and XDR — firmware, MFT tools, and driver stack must match this generation.
NOT bundled with QSFP112 transceivers, AOC/DAC cables, low-profile bracket, or accessories.


Capabilities & Features

RDMA: RoCE v1 / v2 (Ethernet) and native InfiniBand verbs (XDR / NDR / HDR / EDR).
NVMe-oF: NVMe over Fabrics target and initiator over RoCEv2 or InfiniBand.
GPUDirect RDMA & GPUDirect Storage: zero-copy data path between NVIDIA GPUs and remote memory / storage.
SHARPv4 In-Network Computing: on-fabric collectives acceleration for AI training (InfiniBand mode).
ASAP²: hardware-accelerated OVS / OVS-DPDK offload, conntrack offload, VXLAN/Geneve encap-decap.
SR-IOV & PCIe Multi-Host for virtualization and up to 4 independent hosts per adapter.
Inline crypto offloads: IPsec, MACsec, TLS / TLS 1.3 at line rate.
Programmable Congestion Control and Adaptive Routing for AI/HPC fabrics.
Precision timing: PTP IEEE 1588v2, SyncE, hardware timestamping.


Compatibility & Constraints

Drivers: NVIDIA MLNX_OFED / DOCA-Host (Linux), in-box mlx5_core (recent upstream Linux), WinOF-2 (Windows Server), nmlx5 (VMware ESXi). Use a release that explicitly lists ConnectX-8 support.
Host slot: PCIe x16 required. For full 400 Gb/s line rate use a PCIe Gen5 x16 or Gen6 x16 slot. Gen4 x16 will function but caps host bandwidth below 400 Gb/s sustained.
Compatible switches — InfiniBand (XDR / NDR): NVIDIA Quantum-X800 (Q3200 / Q3400 series) and Quantum-2 (QM9700 / QM9790) at NDR400.
Compatible switches — Ethernet (400GbE): NVIDIA Spectrum-4 SN5400 / SN5600 and Spectrum-X platforms.
Compatible interconnect: QSFP112 400G DAC / AOC / transceivers (e.g., MMA4Z00-NS400). For OSFP-cage switches, use QSFP112-to-OSFP fly-over / breakout cables.
Power & cooling: 400G/XDR NICs dissipate more than older ConnectX generations — verify server airflow (typically ≥ 200 LFM).


Frequently Asked Questions (FAQ)
Q. What is the NVIDIA Mellanox 900-9X81Q-00SN-ST0?
It is a ConnectX-8 SuperNIC — a single-port QSFP112 PCIe Gen6 x16 adapter supporting 400GbE (Ethernet) or XDR400 / NDR 400 Gb/s InfiniBand in VPI dual-protocol mode.
Q. Is 900-9X81Q-00SN-ST0 single-port or dual-port?
Single-port QSFP112. Dual-port and 800 Gb/s variants use different MPN suffixes (e.g., 00BN, 00CN).
Q. Will this card run at full 400 Gb/s in a PCIe Gen4 slot?
It will link, but PCIe Gen4 x16 caps host bandwidth around 256 Gb/s usable, below the 400 Gb/s line rate. Use PCIe Gen5 x16 or Gen6 x16 for sustained full-rate workloads.
Q. What drivers does this adapter require?
NVIDIA MLNX_OFED or DOCA-Host on Linux, in-box mlx5_core on recent kernels, WinOF-2 on Windows, nmlx5 on VMware ESXi. Use a release that lists ConnectX-8 support.
Q. Which switches and cables are compatible with 900-9X81Q-00SN-ST0?
Ethernet: NVIDIA Spectrum-4 SN5400 / SN5600 with 400G QSFP112 DAC, AOC, or transceivers. InfiniBand: Quantum-X800 or Quantum-2 (at NDR400) with QSFP112 cables; QSFP112-to-OSFP breakouts available for OSFP-cage switches.
Q. Is this the same as a BlueField DPU?
No. The ConnectX-8 SuperNIC is a fixed-function RDMA NIC. BlueField-3 / BlueField-4 are DPUs with onboard Arm cores running Linux and DOCA services — a different product category.
Q. What condition is supplied by T.E.S IT-SOLUTIONS?
Refurbished Grade A. Each unit is bench-tested by T.E.S IT-SOLUTIONS — PCIe link establishment, port link verification, EEPROM / vendor ID check, and full cosmetic inspection. EU stock, anti-static packaging, global B2B shipping.

Why buy from T.E.S IT-SOLUTIONS
T.E.S IT-SOLUTIONS is a Europe-based NVIDIA Mellanox specialist serving HPC, AI, and enterprise data-center procurement teams since 2007. Every unit is identified by MPN, NVIDIA P/N, and ASIC generation before it ships, and every customer receives free pre-sale compatibility advice for switch, adapter, transceiver, and cable pairing. Multi-currency B2B invoicing, EU stock with worldwide express options, and direct technical contact with our engineering team — not a call-center.