NVIDIA ConnectX-6 InfiniBand Adapters

NVIDIA ConnectX-6 InfiniBand Adapters

High-Speed, Scalable, and Intelligent Network Connectivity for HPC, AI, and Cloud Workloads

In today’s data-driven world, high-performance computing (HPC), artificial intelligence (AI), and large-scale cloud infrastructures require more than just fast processors.

Efficient networking plays a crucial role in ensuring applications run smoothly, reliably, and at scale. NVIDIA ConnectX-6 InfiniBand Adapters deliver exactly that.

Designed as part of the NVIDIA Quantum InfiniBand platform, ConnectX-6 adapters are engineered to handle the most complex workloads with unprecedented speed, reliability, and flexibility.

These smart network interface cards (NICs) offer up to two ports of 200Gb/s InfiniBand and Ethernet connectivity, combined with NVIDIA’s unique In-Network Computing acceleration technologies.

Key Benefits and Advantages

• Up to 200Gb/s InfiniBand and Ethernet connectivity for maximum throughput
• Support for both PCIe 3.0 and PCIe 4.0 interfaces for optimal compatibility
• In-Network Computing acceleration for HPC, AI, and cloud applications
• Ultra-low latency and high message rates ideal for real-time data-intensive tasks
• Advanced offload capabilities to reduce CPU load and enhance overall system efficiency
• VPI (Virtual Protocol Interconnect) support for InfiniBand and Ethernet in one adapter
• Scalable architecture suitable for future data center growth

Core Features

1. Unmatched Performance with 200Gb/s Connectivity

ConnectX-6 adapters deliver up to 200 gigabits per second (Gb/s) data rate, supporting both InfiniBand HDR and 200GbE Ethernet protocols.

This enables data centers to handle massive workloads such as scientific simulations, large-scale AI model training, and hyperscale storage operations effortlessly.

2. Dual and Single Port Options

Depending on infrastructure needs, customers can choose from single or dual port configurations.

Dual port cards are especially beneficial in environments where high availability, failover, or increased bandwidth is required.

3. PCI Express 4.0 and 3.0 Compatibility

ConnectX-6 adapters support both PCIe 4.0 and PCIe 3.0 buses. This ensures compatibility with a wide range of server architectures, while also unlocking the full potential of newer systems that take advantage of PCIe 4.0’s higher bandwidth.

4. In-Network Computing Acceleration

NVIDIA’s In-Network Computing technology allows critical data operations to happen directly within the network, rather than at the server endpoint.

This reduces data movement overhead, decreases latency, and dramatically improves workload efficiency for distributed applications in HPC, AI, and cloud services.

5. Smart Offloads and RDMA Support

By offloading complex networking tasks from the CPU to the ConnectX-6 adapter, system resources are freed up to handle core application processing.

Remote Direct Memory Access (RDMA) support ensures ultra-low latency data transfers across servers without consuming CPU resources, making it perfect for latency-sensitive applications.

6. VPI (Virtual Protocol Interconnect)

ConnectX-6 is equipped with VPI technology, allowing seamless switching between InfiniBand and Ethernet modes.

This dual-protocol capability provides unmatched flexibility for hybrid network environments.

Detailed Specifications

Feature	Description
Product Name	NVIDIA ConnectX-6 InfiniBand Adapters
Data Rate	100GbE / 200GbE
Ports	1 or 2 (QSFP56)
PCIe Interface	PCIe 3.0 x16, PCIe 4.0 x8/x16
Network Interface	QSFP56
Technology	InfiniBand HDR / EDR and 100GbE Ethernet
Offload Capabilities	RDMA, GPU Direct, SR-IOV, In-Network Computing
Compatibility	Compatible with all major server vendors supporting PCIe 3.0/4.0
Security	Hardware isolation, secure boot, and cryptographic features (varies by SKU)

Product Models and Options

SKU	Model	Ports	PCIe	Data Rate	Crypto
900-9X6AF-0018-MT0	MCX653105A-HDAT-SP	Single	PCIe 4.0 x16	200GbE/HDR	–
900-9X6AF-0058-ST2	MCX653106A-HDAT-SP	Dual	PCIe 4.0 x16	200GbE/HDR	–
900-9X6AF-0016-ST2	MCX653105A-ECAT-SP	Single	PCIe 3.0/4.0 x16	HDR100/EDR/100GbE	–
900-9X6AF-0056-MT0	MCX653106A-ECAT-SP	Dual	PCIe 3.0/4.0 x16	HDR100/EDR/100GbE	–
900-9X628-0016-ST0	MCX651105A-EDAT	Single	PCIe 4.0 x8	HDR100/EDR/100GbE	–
900-9X657-0058-SB0	MCX653436A-HDAB	Dual	PCIe 4.0 x16	HDR/200GbE	–

Application Scenarios

• High-Performance Computing (HPC) Clusters
• AI Model Training and Inference at Scale
• Hyperscale Cloud Platforms
• Financial Services Requiring Ultra-Low Latency
• Storage Area Networks (SANs)
• Enterprise Data Centers Supporting Hybrid Workloads

Why Choose NVIDIA ConnectX-6 InfiniBand Adapters?

✔ Industry-Leading Speed and Efficiency: The 200Gb/s throughput combined with In-Network Computing creates unparalleled network acceleration.

✔ Future-Proof Compatibility: PCIe 4.0 support ensures your infrastructure investment remains relevant for years to come.

✔ Flexibility Across Protocols: With VPI, ConnectX-6 provides InfiniBand and Ethernet in a single adapter, simplifying deployment and scaling.

✔ Reduced TCO: Smart offloads and RDMA reduce server CPU usage, lowering power consumption and operational costs over time.

✔ Proven Reliability: NVIDIA’s engineering quality and global support make ConnectX-6 a trusted choice for mission-critical applications.

Conclusion

The NVIDIA ConnectX-6 InfiniBand Adapters represent a new benchmark in smart networking for HPC, AI, and cloud environments.

Combining blazing-fast speeds, advanced offload capabilities, and versatile configuration options, ConnectX-6 enables organizations to scale their computing power efficiently while maintaining reliability and control.

Whether building a next-generation AI cluster or upgrading an existing HPC system, ConnectX-6 delivers the performance and flexibility modern data centers demand.