China Optical Transceiver Module manufacturer
Hong Kong Starsurge Group Co., Limited
Hong Kong Starsurge Group Co., Limited
4
Home > Products > Mellanox Network Card >

NVIDIA ConnectX-7 NDR 400Gb/s InfiniBand Adapter MCX75310AAS-HEAT | PCIe Gen5 Dual-Port

NVIDIA ConnectX-7 NDR 400Gb/s InfiniBand Adapter MCX75310AAS-HEAT | PCIe Gen5 Dual-Port

Brand Name Mellanox
Model Number MCX75310AAS-HEAT(900-9X766-003N-ST0)
Minimum Order Quantity 1pcs
Price Negotiate
Payment Terms T/T
Supply Ability Supply by project/batch
Delivery Time Based on inventory
Packaging Details outer box
Model NO. MCX75310AAS-HEAT(900-9X766-003N-ST0)
Ports Single-Port
Technology Infiniband
Interface Type Osfp56
Specification 16.7cm x 6.9cm
Origin India / Israel / China
Transmission Rate 200gbe
Host Interface Gen3 X16
Detailed Product Description
NVIDIA ConnectX-7 NDR 400Gb/s InfiniBand Smart Adapter
MCX75310AAS-HEAT | PCIe Gen5 Dual-Port 400GbE / NDR

Ultra-low latency RDMA networking for AI factories, HPC clusters, and hyperscale cloud. Supports InfiniBand NDR (400Gb/s) and Ethernet up to 400GbE. In-Network Computing engines, hardware security acceleration, and GPUDirect® Storage ready.

✓ PCIe 5.0 x16 ✓ RoCE & InfiniBand ✓ Secure Boot + IPsec/TLS/MACsec inline ✓ SHARPv3 collective offloads
Product Overview

The NVIDIA ConnectX-7 family redefines data center interconnectivity with up to 400Gb/s bandwidth, hardware-based reliable transport, and advanced in-network computing. The model MCX75310AAS-HEAT belongs to the ConnectX-7 VPI adapter series supporting both InfiniBand (NDR/HDR/EDR) and Ethernet (400GbE to 10GbE). Built for AI, scientific computing, and modern software-defined storage, it delivers sub‑microsecond latency and frees CPU cores via acceleration engines for security, virtualization, and NVMe‑oF. With PCIe Gen5 host interface, this adapter maximizes data movement for GPU clusters and high‑frequency trading environments.

Leveraging ASAP² (Accelerated Switch and Packet Processing) and Zero-Touch RoCE, the adapter simplifies deployment while providing inline hardware encryption – from edge to core. Enterprises can future-proof their infrastructure with NDR speeds and flexible multi‑host support.

Key Features
 Ultra-high throughput

Dual-port NDR 400Gb/s InfiniBand or 400GbE, PCIe Gen5 x16 (up to 32 lanes).

 In‑Network Computing

SHARPv3 collective operations offload, rendezvous protocol offload, on‑board memory for bursts.

 Hardware Security Engines

Inline IPsec/TLS/MACsec AES-GCM 128/256-bit, secure boot with hardware root-of-trust, flash encryption.

 Precision Timing

IEEE 1588v2 PTP (12ns accuracy), SyncE, class C G.8273.2, time-triggered pacing.

 Accelerated Storage & GPUDirect

GPUDirect RDMA, GPUDirect Storage, NVMe‑oF offload, T10-DIF signature handover.

 SDN & Overlay offloads

VXLAN, GENEVE, NVGRE, programmable flexible parser, connection tracking (L4 firewall).

Technology – Built for Data‑Driven Acceleration

NVIDIA ASAP² (Accelerated Switch and Packet Processing) offloads entire data paths, delivering line‑rate SDN performance with zero CPU overhead. ConnectX‑7 integrates advanced memory virtualization: On‑Demand Paging (ODP), User Mode Registration (UMR), and address translation services (ATS) for efficient containerized workloads. With support for SR-IOV, VirtIO acceleration, and Multi‑Host technology (up to 4 hosts), the adapter enables bare‑metal performance in virtualized environments. The hardware RoCE engine ensures zero‑touch deployment for converged Ethernet fabrics, while the in‑network memory reduces latency for MPI and collective communication libraries (NVIDIA HPC-X, UCX, NCCL).

To address security from edge to core, inline cryptography engines handle IPsec, TLS 1.3, and MACsec without consuming execution cores, making it ideal for multi‑tenant clouds and financial services.

Typical Deployments
  • AI & Large Language Model clusters: RDMA & GPUDirect accelerating distributed training across hundreds of GPUs.
  • High‑Performance Computing (HPC): MPI offloads, SHARPv3 for in‑network reduction, adaptive routing.
  • Cloud and Edge data centers: Overlay network acceleration, VXLAN offload, SR-IOV.
  • Software‑defined storage (SDS): NVMe/TCP, NVMe‑oF, iSER, NFS over RDMA.
  • Real‑time analytics & trading: Ultra-low latency + PTP class C synchronization.
Compatibility
Component / EcosystemSupported Options
Host InterfacePCIe Gen5 (up to x32 lanes, x16 typical), PCIe bifurcation, Multi-Host (4 hosts)
Operating SystemsLinux (RHEL, Ubuntu, in-box drivers), Windows Server, VMware ESXi (SR-IOV), Kubernetes
Virtualization & ContainersSR-IOV, VirtIO acceleration, NVMe over TCP offload for containers
HPC Middleware / LibrariesOpenMPI, MVAPICH, MPICH, UCX, NCCL, UCC, OpenSHMEM, PGAS
Storage ProtocolsNVMe‑oF (RDMA/TCP), SRP, iSER, NFS over RDMA, SMB Direct, GPUDirect Storage
Technical Specifications
ParameterSpecification
Product ModelMCX75310AAS-HEAT
Form FactorsPCIe HHHL / FHHL, OCP 3.0 SFF/TSFF (refer to ordering guide)
InfiniBand SpeedsNDR 400Gb/s, HDR 200Gb/s, EDR 100Gb/s
Ethernet Speeds400GbE, 200GbE, 100GbE, 50GbE, 25GbE, 10GbE (NRZ/PAM4)
Network Ports1/2/4 port configurations (dependent on SKU; 2-port typical for this model)
Host BusPCI Express Gen 5.0 (compatible with Gen4/Gen3), up to x32 lanes, TLP processing hints, ATS, PASID
Security OffloadsInline IPsec (AES-GCM 128/256), TLS 1.3, MACsec, secure boot, flash encryption, device attestation
RDMA / TransportHardware reliable transport, XRC, DCT, On-Demand Paging, UMR, GPUDirect RDMA
Timing & SyncIEEE 1588v2 PTP, G.8273.2 Class C, 12 ns accuracy, SyncE, PPS in/out, time-triggered scheduling
ManagementNC-SI, MCTP over PCIe/SMBus, PLDM (DSP0248/0267/0218), SPDM, SPI flash, JTAG
Remote BootInfiniBand remote boot, iSCSI, UEFI, PXE
Open NetworkingASAP² for SDN, VXLAN/GENEVE/NVGRE offload, flow mirroring, hierarchical QoS, L4 connection tracking
Power & EnvironmentalRefer to user manual (typical ~15W-25W depending on configuration), operating temp: 0°C to 55°C (not publicly specified, confirm per SKU)
Note: Full specifications vary by exact SKU. Please confirm port count and cable support with official datasheet or sales team.
Selection Guide – ConnectX-7 VPI Adapters
Part Number ExamplePorts / SpeedForm FactorTypical Use
MCX75310AAS-HEATNDR 400Gb/s dual-port, PCIe 5.0 x16PCIe HHHL / FHHLTop‑of‑rack AI & HPC, GPU clusters
MCX75340AAS-HEAT (example)4‑port NDR200 / 200GbEOCP 3.0Hyperscale & multi-host
Other ConnectX-7 SKUsSingle/dual port, 100G/200G/400GOCP, FHHLEnterprise, storage appliances

For assistance with choosing correct SKU (cable support, bracket type, tall/short bracket) contact Starsurge team.

Advantages – Why Choose Starsurge + NVIDIA ConnectX-7
  • Genuine & certified parts – full warranty, secure supply chain.
  • Pre‑sales architecture consulting – HPC / AI cluster design, RDMA tuning.
  • Global logistics & fast fulfillment – stocks in Asia/Europe/Americas.
  • Custom firmware & OEM options – request bulk configuration.
  • Post‑sales technical support – driver integration, firmware updates, RMA.
Service & Support

Hong Kong Starsurge Group provides full lifecycle support: from compatibility validation to deployment and ongoing maintenance. Our engineers assist with:

  • InfiniBand fabric design and performance optimization.
  • RoCE configuration for Ethernet environments.
  • Integration with GPUDirect Storage and NVMe‑oF targets.
  • 3-year standard warranty plus extended coverage available.

Volume pricing, bulk shipping, and customization (brackets, pull-tabs) are available upon request.

Frequently Asked Questions
1. What is the primary difference between NDR InfiniBand and 400GbE mode?

NDR includes advanced in-network computing features like SHARP, adaptive routing, and hardware reliable transport natively; Ethernet mode uses RoCE for RDMA with the same line rate. ConnectX-7 supports both via dual personality.

2. Is the MCX75310AAS-HEAT compatible with PCIe Gen4 motherboards?

Yes, the adapter is backward compatible with PCIe Gen4 and Gen3, but at reduced lane speed. Best performance is achieved on PCIe Gen5 slots.

3. Does this model support Multi‑Host?

Yes, ConnectX-7 supports NVIDIA Multi-Host technology, allowing up to 4 independent hosts to share the same adapter (specific SKU and cabling dependent). Please confirm with ordering guide.

4. What encryption standards are accelerated inline?

IPsec (AES-GCM 128/256-bit), TLS (AES-GCM 128/256-bit), and MACsec (AES-GCM 128/256-bit). All offloaded from CPU.

5. Can I use this adapter for NVMe/TCP acceleration?

Absolutely. ConnectX-7 offloads NVMe over TCP as well as NVMe‑oF, reducing host CPU overhead for storage.

Precautions & Installation Notes
  • Ensure proper airflow in server chassis when using high-power transceivers (active optical or DAC).
  • Use firmware provided by NVIDIA or validated by Starsurge for security fixes.
  • Install the latest driver (MLNX_OFED or native inbox) to enable full hardware offloads.
  • Verify QSFP-DD or OSFP compatibility for 400G optics; refer to NVIDIA ethernet transceiver guide.
  • Anti-static handling required – the adapter bears sensitive PCIe components.
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 customers worldwide with products including network switches, NICs, wireless access points, controllers, cables, and related equipment. Backed by an experienced sales and technical team, Starsurge supports industries such as government, healthcare, manufacturing, education, finance, and enterprise. The company also offers IoT solutions, network management systems, custom software development, multilingual support, and global delivery. With a customer-first approach, Starsurge focuses on reliable quality, responsive service, and tailored solutions that help clients build efficient, scalable, and dependable network infrastructure.

✓ Authorized partner for leading brands including NVIDIA networking products.

Key Facts – At a Glance
Max Throughput400 Gb/s (NDR / 400GbE)PCIe Gen5Up to 32 lanes
RDMA SupportInfiniBand & RoCESecurityIPsec/TLS/MACsec inline
GPUDirectRDMA + StorageTiming Accuracy12ns PTP
Compatibility Matrix – Quick Reference
ComponentRecommended / SupportedRemarks
GPU Servers (HGX/DGX)Fully compatible with GPUDirectSHARP offload on InfiniBand
Top-of-Rack SwitchesNVIDIA QM9700 (NDR), SN4000 (Ethernet)Supported
Virtualization hostsVMware ESXi 7.0/8.0, KVM, Hyper-VSR-IOV enabled
Storage SystemsPureStorage, VAST Data, WEKA, Dell PowerScaleNVMe-oF / RoCE ready
Cables / OpticsNVIDIA passive copper DAC, active optical (400G SR4/DR4)Confirm compatibility matrix
Buyer Checklist – MCX75310AAS-HEAT
  • Confirm port configuration (2-port NDR or split 2x200G) matches your switch plan.
  • Confirm PCIe slot physical length and bracket type (HHHL / FHHL).
  • Check airflow direction and passive/active cable requirements.
  • Ensure OS driver support (MLNX_OFED version for your kernel).
  • For security features: enable secure boot & inline crypto in firmware.
  • Review warranty and advance replacement options.
Related Products
NVIDIA Quantum-2 NDR Switch

64-port 400G InfiniBand, SHARPv3 integration.

ConnectX-6 HDR 200G Adapter

Cost‑effective 200Gb/s for smaller clusters.

NVIDIA BlueField-3 DPU

Storage & security acceleration + programmable data path.

400G DAC / AOC cables

Compliant OSFP to 2xOSFP or 2xQSFP56.

Related Products
Email to this supplier
 
From:
Enter your Email please.
To: Hong Kong Starsurge Group Co., Limited
Subject:
Message:
Characters Remaining: (80/3000)