 
|
MCX755106AS-HEAT NVIDIA ConnectX-7 400GbE/NDR SmartNIC – GPUDirect, RoCE, In-line Security
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 400GbE SmartNIC,ConnectX-7 SmartNIC with GPUDirect,NVIDIA SmartNIC with in-line security |
||
Product Description
NVIDIA ConnectX-7 400GbE/NDR SmartNIC
NVIDIA ConnectX‑7 MCX755106AS‑HEAT 400Gb/s Dual‑Port OSFP Adapter
InfiniBand NDR / Ethernet 400GbE · PCIe Gen5 x16 · In‑line IPsec/TLS/MACsec · GPUDirect® · RoCE · Secure Boot
InfiniBand NDR / Ethernet 400GbE · PCIe Gen5 x16 · In‑line IPsec/TLS/MACsec · GPUDirect® · RoCE · Secure Boot
Max throughput: 400Gb/s (2*200Gb/s or 1*400Gb/s)
Host interface: PCIe Gen5 x16 (up to 32 lanes)
Form factor: PCIe HHHL (half‑height, half‑length)
Security engines: IPsec, TLS, MACsec (AES‑GCM 128/256)
In‑Network Computing: SHARP, collective offloads, DCT, XRC
Timing: IEEE 1588v2 (Class C), SyncE, PPS
Host interface: PCIe Gen5 x16 (up to 32 lanes)
Form factor: PCIe HHHL (half‑height, half‑length)
Security engines: IPsec, TLS, MACsec (AES‑GCM 128/256)
In‑Network Computing: SHARP, collective offloads, DCT, XRC
Timing: IEEE 1588v2 (Class C), SyncE, PPS
Product Overview
The NVIDIA ConnectX‑7 MCX755106AS‑HEAT is a dual‑port 400Gb/s SmartNIC that supports both InfiniBand (NDR) and Ethernet (400GbE) protocols. It delivers ultra‑low latency, high throughput, and a rich set of hardware‑accelerated networking, storage, and security features. Designed for the most demanding AI, scientific computing, and hyperscale cloud workloads, ConnectX‑7 offloads critical data‑center functions from the CPU, reducing total cost of ownership while increasing performance and security.
This adapter integrates advanced timing (PTP/SyncE), in‑line encryption (IPsec, TLS, MACsec), and NVIDIA In‑Network Computing engines including SHARP and collective operation offloads. It is the ideal choice for building agile, secure, and high‑performance infrastructures from edge to cloud.
Key Features
- Dual‑port 400Gb/s (NDR / 400GbE)
- PCIe Gen5 x16 (up to 32 lanes)
- InfiniBand 1.5 & Ethernet (RoCE) support
- In‑line IPsec, TLS, MACsec (AES‑GCM 128/256)
- GPUDirect® RDMA & Storage
- NVIDIA SHARP (Scalable Hierarchical Aggregation)
- ASAP2 SDN acceleration (VXLAN, GENEVE, NVGRE)
- IEEE 1588v2 Class C (12 ns accuracy) & SyncE
- NVMe‑oF, NVMe/TCP, T10‑DIF offloads
- Secure boot with hardware root‑of‑trust
Technology & Acceleration Engines
NVIDIA In‑Network Computing
ConnectX‑7 integrates collective communication offloads (SHARP), rendezvous protocol offload, and on‑board memory to accelerate MPI and deep learning frameworks. Combined with GPUDirect RDMA, data can move directly between GPU memory and the network, bypassing the CPU.
ASAP2 (Accelerated Switch and Packet Processing)
Delivers line‑rate SDN and VNF acceleration with overlay networking (VXLAN, GENEVE, NVGRE), flexible parser, connection tracking, and flow mirroring – all without CPU overhead.
Security Offloads
Hardware engines for IPsec, TLS, and MACsec with AES‑GCM 128/256‑bit keys, plus platform security (secure boot, flash encryption, device attestation).
Precision Timing
Advanced PTP (IEEE 1588v2) with 12 ns accuracy, G.8273.2 Class C, SyncE, and configurable PPS in/out, ideal for 5G, financial trading, and distributed data centers.
Typical Deployments
- AI & Deep Learning – GPU‑accelerated training and inference with GPUDirect RDMA and SHARP
- High‑Performance Computing (HPC) – MPI workloads, scientific simulations, and weather modeling
- Hyperscale Cloud – SDN, network function virtualization (NFV), and overlay networks
- Enterprise Storage – NVMe‑oF, RoCE, and GPUDirect Storage for fast data access
- Financial Services – low‑latency trading with precise PTP timing and MACsec security
Compatibility
InfiniBand NDR Ethernet 400G/200G/100G/50G/25G/10G RoCE
Supports dual‑protocol operation (InfiniBand and Ethernet) with auto‑negotiation. Compatible with NVIDIA Quantum‑2 switches, Spectrum‑4 switches, and industry‑standard 400GbE platforms. Host interface: PCIe Gen5 (x16) with support for bifurcation, Multi‑Host (up to 4 hosts), SR‑IOV, ATS, PASID, and ACS.
Supports dual‑protocol operation (InfiniBand and Ethernet) with auto‑negotiation. Compatible with NVIDIA Quantum‑2 switches, Spectrum‑4 switches, and industry‑standard 400GbE platforms. Host interface: PCIe Gen5 (x16) with support for bifurcation, Multi‑Host (up to 4 hosts), SR‑IOV, ATS, PASID, and ACS.
Operating Systems – In‑box drivers for Linux (RHEL, Ubuntu), Windows, VMware ESXi (SR‑IOV), and Kubernetes.
Specifications
| Parameter | Details |
|---|---|
| Product model | MCX755106AS‑HEAT |
| Network protocols | InfiniBand (NDR/HDR/EDR) · Ethernet (400/200/100/50/25/10GbE) |
| Number of ports | 2 (OSFP) |
| Host interface | PCIe Gen5 (x16) · up to 32 lanes · bifurcation support |
| Form factor | PCIe HHHL (half‑height, half‑length) |
| Security offloads | IPsec, TLS, MACsec (AES‑GCM 128/256) · secure boot · flash encryption |
| Timing & synchronization | IEEE 1588v2 (Class C, 12 ns) · SyncE · PPS in/out |
| Storage offloads | NVMe‑oF, NVMe/TCP, SRP, iSER, NFS over RDMA, T10‑DIF |
| SDN acceleration | ASAP2: VXLAN, GENEVE, NVGRE, L4 firewall, flow mirroring, QoS |
| In‑Network Computing | SHARP, collective offloads, rendezvous offload, on‑board memory |
| Virtualization | SR‑IOV, VirtIO acceleration, Multi‑Host (4 hosts) |
| Management | NC‑SI, MCTP, PLDM (monitor, firmware, FRU), SPDM, SPI flash, JTAG |
| Remote boot | IB boot, iSCSI, UEFI, PXE |
Selection Guide
The ConnectX‑7 family offers various form factors and port configurations. The MCX755106AS‑HEAT is a dual‑port HHHL adapter with OSFP connectors. For OCP 3.0 or other form factors, please refer to the ConnectX‑7 user manuals.
| Ordering Part Number | Ports | Form Factor | Host Interface | Description |
|---|---|---|---|---|
| MCX755106AS‑HEAT | 2x OSFP | PCIe HHHL | PCIe Gen5 x16 | 400Gb/s dual‑port, InfiniBand/Ethernet, full security & timing |
| MCX755106AS‑HEBT | 2x OSFP | PCIe HHHL | PCIe Gen5 x16 | Same, with different cooling/backplate options |
| MCX755106AS‑HEAT‑OCP | 2x OSFP | OCP 3.0 | PCIe Gen5 x16 | OCP 3.0 form factor |
Contact Starsurge for volume pricing, custom SKUs, and lead time.
Advantages
- Ultra‑low latency – sub‑microsecond latency for HPC and AI workloads
- CPU offload – hardware engines for networking, storage, and security free up CPU cores
- Unified fabric – single adapter for InfiniBand and Ethernet, reducing inventory complexity
- Future‑proof – PCIe Gen5 and 400Gb/s readiness for next‑gen data centers
- Enhanced security – in‑line encryption at line rate with zero performance penalty
- Precision timing – Class C PTP and SyncE for time‑sensitive applications
Service & Support
Hong Kong Starsurge Group offers comprehensive support for NVIDIA ConnectX‑7 adapters:
- Technical consultation for system integration and performance tuning
- Global logistics with fast delivery from regional hubs
- Firmware update and secure boot configuration assistance
- 3‑year warranty with advanced replacement
FAQ
Q: What is the difference between InfiniBand and Ethernet modes on this adapter?
A: The adapter supports both protocols and can be configured via firmware. InfiniBand mode offers RDMA, SHARP, and hardware transport; Ethernet mode supports RoCE, overlay networking, and standard L2/L3 features.
Q: Does it support GPUDirect?
A: Yes, GPUDirect RDMA and GPUDirect Storage are fully supported, enabling direct data movement between GPU and network.
Q: Can I use this adapter in a virtualized environment?
A: Yes, SR‑IOV, VirtIO, and Multi‑Host allow efficient virtualization and container networking.
Q: What is the power consumption?
A: Please confirm with NVIDIA or your Starsurge representative for specific power numbers, as it depends on configuration and usage.
Q: Which PCIe bifurcation modes are supported?
A: The adapter supports various bifurcation modes (x16, x8/x8, x4/x4/x4/x4) depending on the host system.
Handling & Precautions
- Use ESD‑safe practices when handling the adapter – avoid touching the gold fingers
- Ensure adequate airflow in the server chassis for proper cooling
- Do not force the adapter into the PCIe slot; align carefully
- Firmware updates should be performed per NVIDIA guidelines to avoid disruption
- For security features (secure boot, encryption), follow the platform security enablement guide
About Hong Kong Starsurge Group
Hong Kong Starsurge Group Co., Limited is a technology‑driven provider of network hardware, IT services, and system integration solutions. Founded in 2008, we serve customers worldwide with products including network switches, NICs, wireless access points, controllers, cables, and related equipment. With an experienced sales and technical team, Starsurge supports industries such as government, healthcare, manufacturing, education, finance, and enterprise. We also offer IoT solutions, network management systems, custom software development, multilingual support, and global delivery. Our customer‑first approach ensures reliable quality, responsive service, and tailored solutions that help clients build efficient, scalable, and dependable network infrastructure.
Key Facts
- 400Gb/s dual‑port OSFP
- PCIe Gen5 x16 (32 lanes)
- InfiniBand NDR & Ethernet 400GbE
- In‑line IPsec/TLS/MACsec
- GPUDirect RDMA + Storage
- SHARP & collective offloads
- IEEE 1588v2 Class C (12 ns)
- NVMe‑oF, NVMe/TCP offload
- ASAP2 SDN acceleration
- Secure boot + flash encryption
Compatibility Matrix
| System | Port Type | Speed | Protocol |
|---|---|---|---|
| NVIDIA Quantum‑2 | OSFP | 400Gb/s | InfiniBand NDR |
| NVIDIA Spectrum‑4 | OSFP | 400GbE | Ethernet |
| DGX H100 / HGX | OSFP | 400Gb/s | InfiniBand / RoCE |
| Third‑party 400G switches | OSFP/QSFP‑DD | 400GbE | Ethernet (RoCE) |
Buyer Checklist
- ✓ Confirm PCIe slot: Gen5 x16 (or lower Gen with reduced performance)
- ✓ Verify form factor: HHHL (half‑height, half‑length) fits your chassis
- ✓ Check OS/driver support (Linux, Windows, VMware, Kubernetes)
- ✓ Determine required security features (IPsec, TLS, MACsec) and enablement
- ✓ Evaluate timing requirements (PTP Class C, SyncE)
Related Products
- NVIDIA ConnectX‑7 OCP 3.0 adapters (MCX755106AS‑HEAT‑OCP)
- NVIDIA Quantum‑2 / Spectrum‑4 switches
- NVIDIA BlueField‑3 DPUs for advanced data‑center offload
- NVIDIA GPUDirect‑enabled cables and transceivers
Related Guides
- ConnectX‑7 User Manual (PCIe adapters)
- NVIDIA GPUDirect RDMA and Storage documentation
- IEEE 1588v2 & SyncE deployment guide
- ASAP2 SDN acceleration whitepaper
Want to Know more details about this product