900-9X6AG-0056-ST1 Nvidia ConnectX-6 Dx Card 2Port QSFP56 100GBE PCIe 4.0 x16 Ethernet Adapter

Overview of the Nvidia ConnectX-6 Dx Adapter

The Nvidia 900-9X6AF-0016-ST1 ConnectX-6 InfiniBand network adapter is a high-performance 100Gbps single-port QSFP56 PCIe expansion card designed for modern data center environments. Built to support both Ethernet and InfiniBand workloads, this PCIe 3.0/4.0 x16 interface adapter delivers ultra-low latency, exceptional throughput, and scalable networking capabilities for enterprise-grade computing, cloud infrastructure, and HPC systems.

General Information

• Manufacturer: Nvidia
• Part Number: 900-9X6AG-0056-ST1
• Model: ConnectX-6 Dx Dual-Port 100GbE PCIe Adapter
• Crypto Capability: Disabled, Secure Boot Enabled

Technical Specifications

• PCIe Gen3 and Gen4 support with x16 lanes architecture
• High-speed SerDes operation up to 16.0 GT/s
• Backward compatibility with PCIe 2.0 and PCIe 1.1 platforms
• Optimized for stable and scalable data transmission performance
• Operating voltage: 3.3V auxiliary supply
• Maximum current requirement: 100mA

Ethernet Speed

• 100Gb/s: 100GBASE-CR2, CR4, KR4, SR4, KR2, SR2
• 50Gb/s: 50GBASE-R2, R4
• 40Gb/s: 40GBASE-CR4, KR4, SR4, LR4, ER4, R2
• 25Gb/s: 25GBASE-R standard support
• 20Gb/s: 20GBASE-KR2
• 10Gb/s: 10GBASE-LR, ER, CX4, CR, KR, SR
• 1Gb/s: 1000BASE-CX, KX compatibility
• Legacy Mode: SGMII protocol support included

Cooling and Airflow

• Passive cable configuration: 550 LFM airflow (Heatsink to Port, Hot Aisle)
• Active 2.5W cable setup: 700 LFM airflow requirement (Heatsink to Port, Hot Aisle)

High-Performance Benefits

• Ultra-high throughput 100GbE connectivity for intensive workloads
• Optimized for low latency real-time computing applications
• Scalable bandwidth support for growing enterprise demands

Deployment Flexibility

• Supports both copper and fiber optic cabling infrastructures
• Wide compatibility with multiple Ethernet speed standards
• PCIe backward compatibility ensures easy system integration
• Designed for continuous 24/7 data center operations
• Secure Boot support for enhanced system protection
• Durable architecture optimized for mission-critical environments

Compatibility

• Compatible with PCI Express x16 slots (Gen 3.0 and Gen 4.0 supported systems)
• Works with Linux-based distributions including enterprise server kernels
• Supports Microsoft Windows Server environments
• Compatible with VMware ESXi virtualization platforms
• Supports InfiniBand and Ethernet-based networking architectures
• Works with QSFP56 optical modules and DAC/AOC cables
• Compatible with Mellanox/NVIDIA networking drivers and toolsets
• Designed for HPC clusters, cloud servers, and data center switches

The Nvidia ConnectX-6 Dx Card 2-Port QSFP56 100GbE PCIe 4.0

The Nvidia ConnectX-6 Dx 2-Port QSFP56 100GbE PCIe 4.0 x16 Ethernet Adapter is a high-performance networking solution designed for modern data-intensive environments. Built for hyperscale data centers, cloud computing infrastructures, enterprise virtualization clusters, and AI-driven workloads, this advanced network interface card delivers ultra-low latency, high throughput, and robust security capabilities. It is engineered to handle demanding workloads such as distributed storage, machine learning training clusters, real-time analytics, and software-defined networking applications.

This adapter belongs to the next generation of intelligent network interface cards that integrate hardware acceleration for networking, security, and storage offload. With dual 100 Gigabit Ethernet ports using QSFP56 interfaces, it enables extremely fast data transfer rates while reducing CPU overhead, making it ideal for high-density server environments.

Architecture and Hardware

The architecture of the ConnectX-6 Dx is built around a highly optimized network processing unit capable of handling millions of packets per second. It leverages advanced silicon design that integrates multiple offload engines for encryption, compression, and transport acceleration. The PCIe 4.0 x16 interface ensures maximum bandwidth between the network adapter and the host system, significantly reducing bottlenecks in high-throughput environments.

The dual-port QSFP56 configuration allows the card to support 100GbE per port, offering a combined potential throughput of 200GbE. This makes it suitable for workloads requiring parallel data streams, redundancy configurations, or load-balanced network architectures. The QSFP56 interface also supports backward compatibility with lower-speed optics such as QSFP28, ensuring flexibility in deployment.

Network Processing Unit (NPU)

At the core of this adapter is a specialized network processing unit designed to handle packet processing tasks independently of the host CPU. This includes routing decisions, encapsulation, tunneling, and packet filtering. By offloading these tasks, the system achieves reduced latency and improved overall compute efficiency.

PCIe 4.0 x16 Interface

The PCIe 4.0 x16 interface provides a theoretical bandwidth of up to 32 GT/s per lane, ensuring that the adapter can fully utilize its 200GbE potential without being constrained by host-side limitations. This is especially important in environments where multiple high-speed NVMe storage arrays and GPU clusters operate simultaneously.

High-Speed Connectivity and QSFP56 Technology

QSFP56 technology is a key feature of this network adapter, enabling 100GbE connectivity per port. QSFP56 transceivers support higher signal integrity and improved thermal efficiency compared to earlier generations. This allows for longer cable runs and more stable connections in dense rack deployments.

The dual-port configuration allows administrators to configure redundancy, link aggregation, or segmentation of traffic between different network layers. This flexibility is essential in modern distributed systems where reliability and scalability are critical.

100GbE Bandwidth Utilization

With 100 Gigabit Ethernet per port, the adapter is capable of handling extremely large data streams such as high-resolution video processing, large-scale database replication, and AI dataset transfers. This bandwidth is particularly useful in machine learning environments where datasets often exceed terabytes in size and require rapid movement between storage and compute nodes.

Advanced Offload Capabilities

One of the most significant advantages of the ConnectX-6 Dx is its extensive offload capabilities. These features reduce CPU utilization and improve system efficiency by handling complex networking tasks directly on the NIC hardware.

RDMA Over Converged Ethernet (RoCE)

Remote Direct Memory Access (RDMA) over Converged Ethernet enables direct memory-to-memory communication between servers without involving the CPU. This drastically reduces latency and increases throughput in distributed systems. The adapter fully supports RoCE v2, making it ideal for high-performance computing clusters and storage networks.

NVMe over Fabrics Acceleration

The adapter provides hardware acceleration for NVMe over Fabrics (NVMe-oF), enabling remote storage devices to be accessed with near-local performance. This is essential for modern data centers that rely on disaggregated storage architectures.

Performance in Data Center Environments

In hyperscale data centers, performance and efficiency are essential. The ConnectX-6 Dx is optimized for high-density deployments where thousands of servers communicate simultaneously. Its ability to handle massive packet rates ensures smooth operation even under peak loads.

The adapter reduces CPU overhead significantly, allowing more compute resources to be allocated to application workloads rather than networking tasks. This improves cost efficiency and overall system performance.

Cloud Computing and Virtualization Use Cases

Cloud environments require highly flexible and scalable networking solutions. The ConnectX-6 Dx supports advanced virtualization technologies that allow multiple tenants to share the same physical hardware securely and efficiently.

NVMe-oF Integration

By supporting NVMe over Fabrics, the adapter enables remote SSDs to be accessed with near-local latency, significantly improving storage performance in virtualized environments.

Deployment Scenarios and Enterprise

Enterprise environments benefit from the versatility of the ConnectX-6 Dx. It can be deployed in financial institutions, telecommunications networks, research facilities, and enterprise data centers.

Its ability to handle mixed workloads makes it suitable for environments where storage, compute, and networking converge on a single infrastructure.