 
|
NVIDIA ConnectX-7 MCX755106AS-HEAT 200G Dual-Port InfiniBand Card – 2× OSFP
Product Details:
| Brand Name: | Mellanox |
| Model Number: | MCX755106AS-HEAT(900-9X7AH-0078-DTZ) |
| Document: | Connectx-7 infiniband.pdf |
Payment & Shipping Terms:
| Minimum Order Quantity: | 1pcs |
|---|---|
| Price: | Negotiate |
| Packaging Details: | outer box |
| Delivery Time: | Based on inventory |
| Payment Terms: | T/T |
| Supply Ability: | Supply by project/batch |
|
Detail Information |
|||
| Model NO.: | MCX755106AS-HEAT(900-9X7AH-0078-DTZ) | Ports: | 2-Port |
|---|---|---|---|
| Technology: | Infiniband | Interface Type: | Osfp56 |
| Specification: | 16.7cm X 6.9cm | Origin: | India / Israel / China |
| Transmission Rate: | 200gbe | Host Interface: | Gen3 X16 |
| Highlight: | Nvidia ConnectX-7 InfiniBand card,200G dual-port network card,OSFP InfiniBand card with warranty |
||
Product Description
NVIDIA ConnectX-7 MCX755106AS-HEAT 200G Dual-Port InfiniBand Card
MCX755106AS-HEAT 200Gb/s InfiniBand Adapter
NDR200 · Dual‑Port OSFP · PCIe 5.0 x16 · Ultra‑Low Latency for HPC & AI
NDR200 · Dual‑Port OSFP · PCIe 5.0 x16 · Ultra‑Low Latency for HPC & AI
Model: MCX755106AS-HEAT
200 Gb/s NDR200 InfiniBand per port
2 * OSFP dual‑port connectivity
400 Gb/s aggregate bandwidth
PCIe 5.0 x16 up to 32 lanes
GPUDirect RDMA & Storage
Product Overview
The NVIDIA MCX755106AS-HEAT is a high-performance dual-port ConnectX-7 InfiniBand host channel adapter (HCA) delivering 200Gb/s NDR200 connectivity per port over two OSFP interfaces, providing an aggregate bandwidth of 400Gb/s. Designed for the most demanding HPC, AI, and enterprise data center environments, this PCIe 5.0 x16 adapter combines ultra-low latency, extreme throughput, and innovative in-network computing engines. With support for RDMA, RoCE, GPUDirect RDMA and Storage, and advanced security offloads, the MCX755106AS-HEAT is the ideal choice for building scalable, lossless InfiniBand fabrics.
Key Features
- Dual-Port NDR200 InfiniBand: Two 200Gb/s OSFP ports with support for NDR200, HDR, EDR, and FDR speeds for backward compatibility.
- PCIe 5.0 x16 Host Interface: Up to 32 lanes with PCIe 5.0 support, delivering maximum bandwidth and low latency to the host system.
- RDMA & RoCE Acceleration: Hardware-offloaded RDMA and RoCE (RDMA over Converged Ethernet) for efficient data movement with zero CPU overhead.
- GPUDirect RDMA & Storage: Direct GPU-to-GPU communication and GPUDirect Storage for accelerated AI/ML and data analytics workflows.
- In-Network Computing: Support for SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) and collective operations offloads.
- Advanced Security Offloads: Inline hardware encryption/decryption for IPsec, MACsec, and TLS with AES-GCM 128/256-bit keys.
- Virtualization & SDN: SR-IOV, VirtIO acceleration, overlay network support (VXLAN, GENEVE, NVGRE), and programmable flexible parser.
- Precision Timing: IEEE 1588v2 PTP with 12 ns accuracy, SyncE, and configurable PPS in/out for time-sensitive applications.
Technology
ConnectX-7 Architecture: The MCX755106AS-HEAT is built on NVIDIA's ConnectX-7 ASIC, which integrates advanced SerDes technology supporting NRZ and PAM4 signaling. It delivers 200Gb/s InfiniBand bandwidth per port over two OSFP interfaces, with hardware offloads for RDMA, RoCE, NVMe-oF, and security protocols. The adapter's in-network computing engines enable SHARP collective acceleration and MPI tag matching, reducing message passing overhead and improving application scalability.
Software-Defined Networking (SDN): NVIDIA ASAP (Accelerated Switch and Packet Processing) technology offloads SDN and VNF acceleration, providing line-rate performance without CPU penalty. The adapter supports overlay networks (VXLAN, GENEVE, NVGRE), connection tracking, and flow mirroring for advanced network telemetry.
Typical Deployments
- High-Density HPC Clusters: Dual-port connectivity enables fat-tree topologies with reduced switch port consumption.
- AI & Machine Learning Training: Enable GPUDirect RDMA for efficient multi-node training of large language models.
- Hyperscale Data Centers: Build scalable, lossless InfiniBand fabrics with high port density and low latency.
- Enterprise Cloud & Storage: Deploy NVMe over Fabrics (NVMe-oF) and RDMA-based storage solutions.
- Telecom & Edge: Leverage precision timing (PTP, SyncE) for 5G and timing-sensitive infrastructure.
Compatibility
| Component / Ecosystem | Compatibility |
|---|---|
| InfiniBand Speeds | NDR 200Gb/s, HDR 200Gb/s, EDR 100Gb/s, FDR 56Gb/s |
| Ethernet Speeds | 200GbE, 100GbE, 50GbE, 25GbE, 10GbE (with RoCE) |
| Host Interface | PCIe Gen5 x16 (up to 32 lanes), backward compatible with PCIe 4.0 |
| Operating Systems | Linux (RHEL, Ubuntu), Windows, VMware ESXi, Kubernetes |
| Software Stacks | MLNX_OFED, HPC-X, UFM, OpenMPI, MVAPICH, MPICH, NCCL, UCX, UCC |
| Security | IPsec, MACsec, TLS, secure boot, flash encryption, device attestation |
Specifications – MCX755106AS-HEAT
| Category | Detail |
|---|---|
| Product Model | MCX755106AS-HEAT |
| Product Family | NVIDIA ConnectX-7 |
| Port Configuration | 2 * OSFP (200Gb/s NDR200 InfiniBand per port, 400Gb/s aggregate) |
| Host Interface | PCIe 5.0 x16 (up to 32 lanes), backward compatible with PCIe 4.0 |
| Supported Protocols | InfiniBand (IBTA Spec 1.5), Ethernet (RoCE) |
| InfiniBand Speeds | NDR 200Gb/s, HDR 200Gb/s, EDR 100Gb/s, FDR 56Gb/s |
| Ethernet Speeds | 200GbE, 100GbE, 50GbE, 25GbE, 10GbE |
| Form Factor | PCIe HHHL, FHHL |
| Interface Technologies | NRZ (10G, 25G), PAM4 (50G, 100G) |
| In-Network Computing | SHARP, MPI Tag Matching, Collective Operations Offloads |
| Security Offloads | IPsec, MACsec, TLS (AES-GCM 128/256-bit), Secure Boot |
| Virtualization | SR-IOV, VirtIO, Overlay Networks (VXLAN, GENEVE, NVGRE) |
| Precision Timing | IEEE 1588v2 (12 ns accuracy), SyncE, PPS in/out |
| Storage Acceleration | NVMe-oF, NVMe/TCP, SRP, iSER, NFS over RDMA, SMB Direct |
| Management Interfaces | NC-SI, MCTP (SMBus, PCIe), PLDM, SPDM, SPI, JTAG |
| Remote Boot | InfiniBand remote boot, iSCSI, UEFI, PXE |
| Operating Conditions | Standard data center environment |
| Regulatory | RoHS, CE, FCC, VCCI (refer to NVIDIA compliance documentation) |
| Included Accessories | Tall bracket |
* Specifications are based on NVIDIA public documentation and may vary. Please confirm details with your sales representative.
Selection Guide
| SKU / Model | Description | Port Speed | Port Count | Interface |
|---|---|---|---|---|
| MCX755106AS-HEAT | ConnectX-7 NDR200 InfiniBand Adapter (Dual-Port) | 200Gb/s per port | 2 * OSFP | PCIe 5.0 x16 |
| MCX75310AAS-HEAT | ConnectX-7 NDR200 InfiniBand Adapter (Single-Port) | 200Gb/s | 1 * OSFP | PCIe 5.0 x16 |
| MCX75310AAS-NEAT | ConnectX-7 NDR400 InfiniBand Adapter (Single-Port) | 400Gb/s | 1 * OSFP | PCIe 5.0 x16 |
| MCX755106AS-NEAT | ConnectX-7 NDR400 InfiniBand Adapter (Dual-Port) | 400Gb/s per port | 2 * OSFP | PCIe 5.0 x16 |
Advantages
- Dual-Port Density: Two 200Gb/s ports in a single PCIe slot reduces server footprint and cabling complexity.
- Ultra-Low Latency: Hardware-offloaded RDMA and in-network compute minimize communication overhead.
- High Throughput: PCIe 5.0 x16 and 400Gb/s aggregate bandwidth deliver extreme performance for data-intensive applications.
- Investment Protection: Supports multiple InfiniBand and Ethernet speeds for mixed-speed environments.
- Enhanced Security: Inline IPsec, MACsec, and TLS offloads secure data in transit without CPU impact.
Service & Support
Hong Kong Starsurge Group Co., Limited provides full lifecycle support for NVIDIA ConnectX-7 adapters:
- Pre-sales consulting: Adapter selection, server compatibility, and fabric design.
- Logistics & delivery: Global shipping with tracking and customs clearance.
- Technical support: Driver installation, firmware updates, and performance tuning.
- Training: Hands-on workshops for RDMA, RoCE, and UFM administration.
- Warranty: Standard 1-year hardware warranty (extended options available).
Frequently Asked Questions
What is the aggregate bandwidth of the MCX755106AS-HEAT?
The adapter provides 400Gb/s aggregate bandwidth (2 * 200Gb/s) across its two OSFP ports.
Is the MCX755106AS-HEAT compatible with PCIe 4.0 servers?
Yes. The adapter is PCIe 5.0 x16 and is backward compatible with PCIe 4.0 slots, although bandwidth may be limited to PCIe 4.0 speeds.
Can this adapter be used with Ethernet switches?
The adapter supports both InfiniBand and Ethernet protocols. When operating in Ethernet mode, it supports RoCE (RDMA over Converged Ethernet) up to 200GbE per port.
What is the difference between MCX755106AS-HEAT and MCX75310AAS-HEAT?
The MCX755106AS-HEAT has two OSFP ports (dual-port) while the MCX75310AAS-HEAT has one OSFP port (single-port). Both support NDR200 speeds.
Does the adapter support GPUDirect?
Yes, it supports GPUDirect RDMA for direct GPU-to-GPU communication and GPUDirect Storage for accelerated data access from storage devices.
Precautions & Safety
- ESD Protection: Use ESD grounding straps when handling the adapter to prevent electrostatic discharge damage.
- Installation: Ensure the server is powered off and disconnected from AC power before installing the adapter.
- Airflow: Ensure adequate airflow around the adapter to prevent overheating. Use the included tall bracket if required.
- Firmware Updates: Follow NVIDIA's recommended procedures for firmware updates to avoid system instability.
- Compatibility: Verify server compatibility and PCIe slot availability before ordering.
Hong Kong Starsurge Group Co., Limited
Founded in 2008, Starsurge is a technology-driven provider of network hardware, IT services, and system integration solutions. We serve government, healthcare, manufacturing, education, finance, and enterprise clients worldwide. With an experienced sales and technical team, we deliver tailored networking solutions — from adapters and switches to IoT and custom software — backed by multilingual support and global delivery. Customer-first, quality-driven, and innovation-focused.
Compatibility Matrix
| NVIDIA Product | Compatible with MCX755106AS-HEAT | Notes |
|---|---|---|
| NVIDIA Quantum-3 Switches (Q3200-RA, Q3400-RA) | ✓ | Native XDR and NDR support |
| NVIDIA Quantum-2 Switches | ✓ | NDR and HDR support |
| NVIDIA DGX / HGX Platforms | ✓ | GPUDirect RDMA and Storage |
| NVIDIA UFM 3.5 | ✓ | Full fabric management and telemetry |
| HPE Servers | ✓ | Standard PCIe slot required |
| Third-party Servers | ✓ | Standard PCIe slot required |
Buyer Checklist
- PCIe Slot: Ensure an available PCIe 5.0 x16 slot (or PCIe 4.0 x16 for backward compatibility).
- Cables/Transceivers: Two OSFP-to-OSFP or OSFP-to-other connector cables for the target switches/NICs.
- Operating System: Verify support for the target OS (Linux, Windows, VMware).
- Driver: Download and install the latest MLNX_OFED or Windows driver from NVIDIA.
- Fabric Management: Plan Subnet Manager (SM) deployment (centralized or distributed).
- Cooling: Ensure adequate server airflow to dissipate heat from the adapter.
Related Products
- NVIDIA Q3400-RA 800G InfiniBand Switch
- NVIDIA Q3200-RA 800G InfiniBand Switch
- OSFP 800G / 400G Transceivers & Cables
- NVIDIA UFM 3.5 Management Appliance
- ConnectX-7 NDR400 Dual-Port Adapter (MCX755106AS-NEAT)
Want to Know more details about this product