Specifications
Ordering Part Number
MCX75310AAS-NEAT
NVIDIA SKU
900-9X766-003N-SQ01
Controller
NVIDIA ConnectX-7
Ports
Single-port OSFP (RHS riding-heatsink cage)
Data Rate
InfiniBand NDR 400 Gb/s (default) / Ethernet 400 GbE
Protocols
InfiniBand: NDR/NDR200/HDR/HDR100/EDR/FDR/SDR · Ethernet: 400/200/100/50/40/10/1 Gb/s
Host Interface
PCIe 5.0 x16 (Gen4/Gen5, 16/32 GT/s SERDES)
Form Factor
PCIe Half Height, Half Length — 68.90 × 167.65 mm
Typical Power
24.9 W with passive cables (PCIe Gen 5.0 x16)
Power Rails
12 V and 3.3 VAUX, 100 mA max auxiliary
Secure Boot
Enabled
Crypto Offload
Disabled on this OPN (see MCX75310AAC-NEAT for crypto enabled)
Bracket
Tall bracket; short bracket included
RDMA
Yes — RoCE and InfiniBand RDMA
SR-IOV
Yes
Operating Temperature
0 °C to 55 °C
Storage Temperature
-40 °C to 70 °C
Operating Systems
Linux, Windows, VMware
MTU
256 B to 4 KB; message sizes up to 2 GB
Lifecycle
Mass production
Overview
MCX75310AAS-NEAT is the single-port 400Gb/s ConnectX-7 adapter in the OSFP form factor, shipped by NVIDIA under SKU 900-9X766-003N-SQ01. It runs InfiniBand NDR at 400 Gb/s by default and can be switched to Ethernet 400GbE, so one card covers both the storage fabric and the management network in an AI cluster node.
The -NEAT suffix indicates Secure Boot enabled and crypto offload disabled — the configuration used for standard AI and storage nodes where IPsec/MACsec is not required. The sibling OPN MCX75310AAC-NEAT is the crypto-enabled part; both are electrically identical apart from that block.
OSFP modules must be fitted in a riding-heatsink cage (RHS) on this adapter, and the card is HHHL so it fits a standard 1U or 2U server riser. Typical power with passive copper cables and a PCIe Gen5 x16 host link is just 24.9 W, which is what makes 400G per node practical in dense GPU servers.
Key Benefits
400G per port: NDR InfiniBand or 400GbE from one card. 24.9 W typical: 400G without a power-budget problem. PCIe 5.0 x16: full line-rate into Xeon 6 / EPYC 9005 hosts.
Applications
GPU cluster fabric interfaces, AI training and inference nodes, storage nodes running NVMe-oF/RDMA, HPC interconnect, and 400G leaf-spine uplinks in AI data centers.
Request a Quote — NVIDIA CONNECTX-7 MCX75310AAS-NEAT — 400G NDR/400GBE SINGLE-PORT OSFP ADAPTER
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →