English 
| Model Number: | |
|---|---|
| Controller: | |
| Host Interface: | |
| Port Configuration: | |
| Ethernet Speed: | |
| Virtualization: | |
| Quantity: | |
MCX516A-CDAT
Mellanox
The MCX516A-CDAT is a dual-port 100GbE Ethernet adapter based on the NVIDIA Mellanox ConnectX-5 controller. It features two QSFP28 connectors and a PCIe 3.0 ×16 host interface, delivering multi-rate Ethernet support from 1GbE up to 100GbE per port. This adapter targets enterprise data centers, cloud infrastructure, and storage networks where standard Ethernet connectivity is required with hardware-accelerated data movement.
Business buyers building scale-out server fleets, virtualization platforms, distributed storage clusters, and AI inference infrastructure can deploy this adapter to increase per-server network capacity while maintaining compatibility with existing Ethernet switching infrastructure.
Parameter | Value |
Part Number | MCX516A-CDAT |
Controller | NVIDIA Mellanox ConnectX-5 |
Host Interface | PCIe 3.0 ×16, 8.0GT/s per lane |
Ports | 2 × QSFP28 |
Ethernet Speeds | 1/10/25/40/50/100GbE |
PHY Types | 100GBASE-CR4, 100GBASE-SR4, 100GBASE-LR4, 50GBASE-R2, 40GBASE-CR4, 25GBASE-CR, 10GBASE-SR/LR |
Protocol Support | Ethernet, RoCE v1/v2 |
Virtualization | SR-IOV, VXLAN, NVGRE offload |
Storage Acceleration | NVMe-oF offload, iSER |
Message Rate (DPDK) | Up to 148 million messages per second |
Typical Power | 12.5W (passive cable configuration) |
Operating Temperature | 0°C to 55°C |
Form Factor | PCIe low-profile bracket |
Compliance | RoHS compliant, CE, FCC |
The adapter supports RDMA over Converged Ethernet (RoCE) v1 and v2, which enables direct memory-to-memory data transfers across Ethernet networks without involving the host CPU in data movement operations. RoCE v2 operates over standard IP routing, allowing deployment across Layer 3 data center fabrics. This capability reduces latency for distributed application traffic and lowers CPU overhead on storage and compute nodes.
Hardware offload for VXLAN and NVGRE tunnel encapsulation and decapsulation allows overlay network traffic to be processed directly on the adapter rather than by the host CPU. This offload maintains line-rate throughput even when virtualized network overlays are in use, supporting cloud and multi-tenant environments where overlay networking is standard. SR-IOV support further enables partitioning of the physical adapter into dedicated virtual functions for direct VM assignment.
Built-in NVMe over Fabrics (NVMe-oF) and iSER acceleration offload storage protocol processing to the adapter silicon. For scale-out storage and hyper-converged infrastructure deployments, this maintains consistent storage access latency while reducing the CPU cycles consumed by storage traffic on each host. Additional stateless offloads include checksum calculation, TSO, LRO, and RSS for standard TCP/IP traffic.
The MCX516A-CDAT fits Ethernet-only data center environments requiring 100GbE per server port. Typical deployments include top-of-rack server uplinks in leaf-spine data center fabrics, distributed storage clusters using NVMe-oF or iSCSI, virtualized private cloud platforms, AI inference clusters with Ethernet-based interconnect, and enterprise backup and replication networks.
Commercial users operating standard Ethernet switching infrastructure can adopt this adapter without requiring InfiniBand-specific switch hardware, simplifying network operations and staff training requirements.
The adapter supports Linux distributions, Windows Server, and VMware ESXi hypervisors. It interoperates with standard QSFP28 optical modules and direct attach copper cables across all supported speed grades. The PCIe 3.0 ×16 interface is backwards compatible with PCIe 2.0 and 1.1 slots at reduced data rates. Driver and firmware updates are available through NVIDIA networking software channels.
Each QSFP28 port supports 1GbE, 10GbE, 25GbE, 40GbE, 50GbE, and 100GbE operation, with auto-negotiation based on the connected transceiver or cable.
Yes. The adapter includes hardware offload for VXLAN and NVGRE tunnel processing, supporting line-rate performance in software-defined and cloud network deployments.
Yes. The adapter supports lower speed grades via breakout cables or compatible transceivers, allowing migration from 25GbE or 40GbE infrastructure to 100GbE without immediate full-fabric replacement.
