NVIDIA ConnectX‑7 400G InfiniBand Adapter MCX75310AAS‑NEAT
NVIDIA Mellanox ConnectX-7 MCX75310AAS-NEAT 400Gb/s NDR InfiniBand Smart Network Adapter engineered for AI factories, scientific computing, and hyperscale cloud. Delivers 400Gb/s throughput, sub-microsecond latency, and hardware-accelerated inline security (IPsec/TLS/MACsec). Supports PCIe Gen5 x16, GPUDirect RDMA, and advanced in-network computing offloads.
Product Overview
The NVIDIA ConnectX-7 family represents a breakthrough in RDMA networking, supporting both InfiniBand and Ethernet protocols up to 400Gb/s. The MCX75310AAS-NEAT model is a high-density single-port NDR InfiniBand adapter designed for demanding HPC and AI clusters. It integrates PCIe Gen5 host interface, hardware offloads for NVMe-oF, collective communications (SHARP), and inline encryption engines, enabling secure, ultra-efficient data centers with lower TCO.
Key Hardware & Performance
- NDR InfiniBand 400Gb/s (compatible with HDR 200G/EDR 100G)
- PCI Express Gen5.0 x16 (supports bifurcation up to x32 lanes)
- Ultra-low latency & hardware-based reliable transport
- GPUDirect RDMA & GPUDirect Storage for GPU-accelerated workloads
- In-Network Computing: SHARPv3 collective offloads
Security & Offload Engines
- Inline IPsec, TLS, MACsec encryption/decryption (AES-GCM 128/256-bit)
- Secure boot with hardware root-of-trust & flash encryption
- ASAP2 (Accelerated Switch and Packet Processing) for SDN
- Connection tracking, flow mirroring, and header rewrite
- Zero-Touch RoCE and overlay network acceleration (VXLAN, GENEVE)
Advanced Technology Accelerators
The ConnectX-7 MCX75310AAS-NEAT incorporates NVIDIA's latest in-network computing engines, offloading critical data path tasks from the CPU:
- RDMA & RoCE: Zero-copy data movement, NVMe-oF offloads (including NVMe/TCP), iSER, SRP, and NFS over RDMA
- Timing & Synchronization: IEEE 1588v2 PTP with 12ns accuracy, SyncE, time-triggered scheduling, PPS input/output
- Virtualization & Multi-Host: SR-IOV, VirtIO acceleration, and NVIDIA Multi-Host technology supporting up to 4 independent hosts
- Data Integrity: T10-DIF signature handover, block-level XTS-AES 256/512 encryption, advanced error reporting (AER)
Typical Deployments & Use Cases
AI & Deep Learning Clusters: Large-scale GPU clusters requiring GPUDirect RDMA and SHARP offloads for NCCL/UCC acceleration.
HPC & Scientific Computing: Weather modeling, molecular dynamics, genome research with MPI libraries and adaptive routing.
Hyperscale Cloud & SDDC: Overlay networking, VXLAN offload, and ASAP2 flow processing for software-defined data centers.
NVMe-oF Storage Arrays: High-performance disaggregated storage leveraging RDMA and hardware NVMe-oF target offload.
Compatibility & Ecosystem
Operating System & Virtualization
In-box drivers for Linux (RHEL, Ubuntu, Debian), Windows Server, VMware ESXi (SR-IOV), and Kubernetes/Container environments.
Compatibility Matrix
| Protocol / Interface |
Supported Speeds / Standard |
| InfiniBand |
NDR (400Gb/s), HDR (200Gb/s), EDR (100Gb/s), FDR (optional) |
| Ethernet (when configured) |
400GbE, 200GbE, 100GbE, 50GbE, 25GbE, 10GbE |
| PCIe Host Interface |
PCIe Gen5.0 x16 (backward compatible to Gen4/Gen3) |
| Security Offloads |
IPsec RoCE, TLS 1.3, MACsec 256-bit |
| Remote Boot |
InfiniBand remote boot, iSCSI, UEFI, PXE |
Technical Specifications
| Parameter |
Specification |
| Product Model |
MCX75310AAS-NEAT |
| Form Factor |
HHHL (Half Height Half Length) |
| Ports |
Single-port NDR InfiniBand (400Gb/s) |
| Host Interface |
PCIe Gen5.0 x16 (up to 32 lanes per Multi-Host option) |
| InfiniBand Standards |
IBTA Spec 1.5 compliant, RDMA, XRC, DCT, adaptive routing |
| Encryption Engines |
Inline IPsec, TLS, MACsec (AES-GCM 128/256-bit) |
| Storage Offloads |
NVMe-oF (NVMe/TCP, NVMe/RDMA), SRP, iSER, T10-DIF |
| Management |
NC-SI, MCTP, PLDM (FRU, Firmware Update, Redfish), SPDM |
| Advanced Timing |
IEEE 1588v2 Class C, 12ns accuracy, SyncE, PTP |
| Secure Boot / RoT |
Hardware root-of-trust, secure firmware update, device attestation |
| Dimensions (typ.) |
68.9mm x 167.6mm (Confirm before ordering) |
Selection Guide: ConnectX-7 Family
When selecting the MCX75310AAS-NEAT, consider port speed (NDR), host slot (PCIe Gen5 x16) and form factor (HHHL). For dual-port or OCP 3.0 requirements, refer to the full ConnectX-7 portfolio:
- Single-port NDR HHHL: MCX75310AAS-NEAT (ideal for GPU servers with limited slots)
- Dual-port NDR / 400G: Check MCX75310AAS-HEAT (confirm ordering guide)
- OCP 3.0 SFF adapter: For rack scale designs
Note: Always verify MPN and firmware region before ordering; datasheet available via NVIDIA docs.nvidia.com/networking
Why Choose MCX75310AAS-NEAT?
Superior TCO: Offloads CPU from networking, security, and storage tasks; reduces server overhead.
Future-ready PCIe Gen5: Doubles bandwidth over Gen4, ideal for next-gen compute nodes.
Zero Trust Security: Line-rate inline crypto with no performance penalty, secure boot ensures infrastructure integrity.
In-network Acceleration: SHARPv3 collective operations reduce MPI traffic, accelerate job completion time.
Service & Support
Hong Kong Starsurge Group Co., Limited provides global logistics, technical consulting, and warranty support. Each adapter undergoes pre-ship validation. Services include:
- 1-year standard warranty with extended options available
- Firmware alignment and configuration assistance for RoCE/NDR deployments
- RMA support via regional hubs (APAC, EMEA, Americas)
- Access to technical documentation and compatibility guides
Frequently Asked Questions
What is the main difference between MCX75310AAS-NEAT and other ConnectX-7 adapters?
This model is a single-port NDR InfiniBand HHHL adapter equipped with full inline security engines and PCIe Gen5 x16 interface, primarily targeting HPC and AI clusters requiring 400G per port.
Does it support Ethernet mode?
Yes, ConnectX-7 supports dual protocol (InfiniBand / Ethernet), but MCX75310AAS-NEAT is typically InfiniBand-focused. For Ethernet-only or dual-protocol need, please consult the VPI ordering guide.
What cooling or power requirements exist?
Passive heatsink design; typical power consumption: Not publicly specified. Please confirm power/thermal requirement for server integration before ordering.
Is this adapter compatible with PCIe Gen4 servers?
Yes, PCIe Gen5 is backward compatible with Gen4 slots, but will operate at Gen4 speeds (max 200Gb/s per port effective bandwidth limited by host).
Precautions & Compliance Notes
- Ensure server motherboard provides PCIe Gen5 x16 electrical and mechanical HHHL slot clearance
- Firmware must be matched to switch fabric (NDR IB switch platform)
- High-speed DAC or active optical cables required for 400G NDR links
- RoHS & REACH compliant; ESD precautions during installation
Company Introduction
Hong Kong Starsurge Group Co., Limited is a technology-driven provider of network hardware, IT services, and system integration solutions. Founded in 2008, the company serves global customers with network switches, NICs, wireless controllers, cables, and full-system networking equipment. Backed by experienced sales and technical teams, Starsurge supports industries including government, healthcare, manufacturing, education, finance, and enterprise. With a customer-first approach, Starsurge delivers tailored solutions, multilingual support, and global delivery to build efficient, scalable network infrastructure.
Key Facts at a Glance
Max Throughput
400Gb/s (NDR)
Inline Crypto
IPsec, TLS, MACsec
Offloads
SHARP, NVMe-oF, ASAP2
PTP Class C
12 ns accuracy
OS Support
Linux, Windows, VMware
Buyer Checklist (Pre-sales)
- Verify server PCIe slot type: HHHL & x16 Gen5/Gen4 compatible
- Confirm InfiniBand fabric speed (400G switch ports required for full NDR)
- Check OS driver availability (in-box or download from NVIDIA)
- Review cooling and power requirements with your chassis spec
- Request quotation with origin warranty terms from Starsurge
Related Products (NVIDIA Mellanox Ecosystem)
- Mellanox Quantum-2 NDR 400G InfiniBand Switch (Q9700)
- Mellanox ConnectX-6 Dx 200G Dual-port EN adapter
- LinkX 400G NDR OSFP DAC / AOC cables
- NVIDIA BlueField-3 DPU for storage and security offload
