NVIDIA MCX75310AAC-NEAT | ConnectX-7 400GbE/NDR Single-Port OSFP Adapter, PCIe 5.0

Q: What is the difference between the MCX75310AAC-NEAT and MCX75310AAS-NEAT models?
A: The MCX75310AAC-NEAT includes hardware cryptographic acceleration enabled, while the MCX75310AAS-NEAT has crypto disabled. Both models support secure boot and offer identical 400Gb/s throughput via a single OSFP port with PCIe 5.0 x16 interface.

Q: Does this adapter support both InfiniBand and Ethernet simultaneously?
A: No, the ConnectX-7 operates as a VPI (Virtual Protocol Interconnect) adapter, meaning it can be configured for either InfiniBand or Ethernet mode, but not both simultaneously. The default mode is InfiniBand NDR, and the mode can be changed through adapter configuration tools.

Q: What GPUDirect features are supported on this adapter?
A: The MCX75310AAC-NEAT supports NVIDIA GPUDirect RDMA for direct memory access between GPUs and the network adapter, and GPUDirect Storage for direct data paths between storage and GPU memory. These features significantly reduce latency and CPU overhead in AI and HPC workloads.

Q: Is this adapter suitable for AI and machine learning workloads?
A: Yes, the ConnectX-7 is specifically designed for AI, machine learning, and HPC applications. With 400Gb/s throughput, advanced RoCE support, GPUDirect capabilities, and hardware-accelerated encryption, it delivers the ultra-low latency and high bandwidth required for distributed GPU training and inference workloads.

Q: What type of OSFP modules are compatible with this adapter?
A: The MCX75310AAC-NEAT requires OSFP modules designed for use with Riding Heat Sink (RHS) cages. The adapter supports standard OSFP flat modules for 400G, 200G, and lower-speed connectivity options compatible with NDR InfiniBand and Ethernet standards.