English 
| Product Model: | |
|---|---|
| Product Series: | |
| NIC Type: | |
| Network Controller: | |
| Host Interface: | |
| Network Ports: | |
| Maximum Transfer Rate: | |
| Quantity: | |
MCX75310AAS-NEAT
Mellanox
The MCX75310AAS-NEAT is a single-port 400Gb/s smart network adapter from the NVIDIA ConnectX-7 VPI family, supporting both Ethernet and InfiniBand protocols through an OSFP form factor. Designed for PCIe 5.0 x16 slots, this adapter delivers high-throughput connectivity for AI training clusters, high-performance computing environments, and GPU-accelerated data center deployments.
Category | Parameter | Value |
General | Product Model | MCX75310AAS-NEAT |
Product Series | NVIDIA Mellanox ConnectX-7 | |
Adapter Type | Smart NIC, Crypto Disabled, Secure Boot Enabled | |
Physical | Form Factor | PCIe Half Height, Half Length |
Dimensions | 68.90mm × 167.65mm (2.71in × 6.6in) | |
Interfaces | Host Interface | PCIe Gen 5.0/4.0/3.0, x16 lanes |
Network Port | Single OSFP port (Riding Heatsink cage only) | |
Performance | InfiniBand Data Rates | NDR/NDR200/HDR/HDR100/EDR/FDR/SDR |
Ethernet Data Rates | 400/200/100/50/40/10/1 Gb/s | |
Typical Power Consumption | 24.9W (passive cables, PCIe Gen 5.0) | |
Voltage | 12V, 3.3VAUX | |
Environmental | Operating Temperature | 0°C to 55°C |
Storage Temperature | -40°C to 70°C | |
Operating Humidity | 10% to 85% relative humidity, non-condensing | |
Operating Altitude | Up to 3050m | |
Regulatory | Safety Certifications | CB, cTUVus, CE |
EMC Certifications | CE, FCC, VCCI, ICES, RCM, KC | |
Environmental Compliance | RoHS Compliant |
The adapter supports RDMA (Remote Direct Memory Access), RoCE (RDMA over Converged Ethernet), and GPUDirect RDMA, which enable direct memory access between network interfaces and GPU memory without CPU intervention. This reduces end-to-end latency and lowers CPU utilization during large data transfers common in distributed AI training workloads.
The ASAP² (Accelerated Switching and Packet Processing) technology provides hardware offload for virtual switching and network functions, supporting overlay networks and network virtualization in cloud and virtualized environments. This offload maintains line-rate throughput while reducing host CPU overhead for software-defined networking deployments.
This adapter is validated for installation in data center servers with proper PCIe slot power delivery and airflow. It is compatible with NVIDIA H100, H200, and B200 GPU server configurations used in large-scale AI clusters. The card is not intended for desktop or workstation installations without adequate cooling and power.
The OSFP port supports passive copper cables, active copper cables, and optical transceivers for both Ethernet and InfiniBand deployments. The adapter supports auto-negotiation with other NVIDIA InfiniBand products for optimized bit error rates and extended cable reach. Backward compatibility includes all previous InfiniBand speeds from NDR down to SDR.
The PCIe slot must supply adequate power for the 24.9W typical operating load. The adapter requires forced airflow consistent with data center server thermal design. Installation in environments without proper airflow may impact device functionality and could void the product warranty.
The InfiniBand implementation complies with IBTA v1.5 specification, with NVIDIA extensions for enhanced performance when connected to other NVIDIA networking products. Ethernet support includes industry-standard IEEE 802.3 protocols from 1GbE through 400GbE.
The MCX75310AAS-NEAT model ships with crypto hardware acceleration disabled, while the MCX75310AAC-NEAT includes crypto acceleration enabled. Both models include Secure Boot functionality and share identical port configurations and performance specifications.
Yes, this is a VPI (Virtual Protocol Interconnect) adapter that supports both protocols. The default configuration boots into InfiniBand mode, and can be configured for Ethernet operation using supported management tools.
PCIe Gen 5.0 x16 provides sufficient host bandwidth for full 400Gb/s line-rate operation. The adapter is backward compatible with PCIe Gen 4.0 and Gen 3.0 slots, though host bandwidth may be constrained in older PCIe generations.
The adapter is supported by standard NVIDIA OFED (OpenFabrics Enterprise Distribution) drivers for Linux distributions, and Windows drivers for applicable server operating systems. Support is included for major enterprise Linux distributions used in data center environments.
