We clean up after ourselves.

The NVIDIA Mellanox ConnectX-6 Dx transforms GPU servers into AI powerhouses with dual-port 100 Gigabit Ethernet connectivity. This PCIe 4.0 network card delivers breakthrough performance for machine learning workloads, providing the bandwidth and ultra-low latency AI training demands.

Built specifically for organizations deploying NVIDIA DGX systems, GPU servers, and AI infrastructure. The ConnectX-6 Dx features hardware-accelerated RoCE for RDMA operations, eliminating CPU overhead and keeping GPUs fed with data during distributed training. Two QSFP56 ports provide 200 Gbps total bandwidth per server.

Perfect for connecting GPU servers to leaf switches in AI clusters. Advanced congestion control prevents packet loss under bursty AI traffic patterns. Hardware-based security offloads TLS and IPsec encryption without performance penalty. Native GPUDirect support accelerates storage and peer-to-peer GPU communication.

Key Benefits:

• Dual 100G ports maximize GPU server bandwidth
• Hardware RoCE offload reduces CPU overhead to near zero
• PCIe 4.0 x16 interface prevents host bottlenecks
• Sub-microsecond latency keeps training synchronized
• GPUDirect RDMA accelerates GPU-to-GPU transfers
• Zero Touch RoCE simplifies deployment
• Hardware encryption offload for TLS and IPsec
• ASAP2 technology accelerates virtual switching

Ideal Use Cases:

• NVIDIA DGX and HGX GPU server connectivity
• AI training clusters requiring low-latency networking
• Machine learning infrastructure with distributed workloads
• High-performance computing and simulation environments
• NVMe over Fabrics storage arrays
• Database servers requiring RDMA acceleration
• Virtualized environments with SR-IOV requirements

Technical Specifications:

• Dual QSFP56 ports supporting 10/25/40/50/100GbE per port
• PCIe 4.0 x16 host interface (works with PCIe 3.0)
• 200 Gbps total bidirectional bandwidth
• Sub-800ns latency for ultra-responsive applications
• 215 million messages per second capability
• Hardware RoCEv2 offload with congestion control
• Stateful connection tracking for L4 firewall offload
• SR-IOV and VirtIO hardware acceleration
• Secure boot with hardware root of trust
• Standard height PCIe bracket included
• Power consumption: 20W typical
• Operating temperature: 0-55C

This refurbished card has been professionally tested and certified to like-new condition. Fully compatible with NVIDIA AI Enterprise, VMware vSphere, Red Hat OpenStack, and major Linux distributions. Works seamlessly with NVIDIA Spectrum switches for complete fabric optimization.

The ConnectX-6 Dx includes advanced telemetry and diagnostic capabilities. Hardware-accelerated histograms provide submicrosecond visibility into queue depths and latency patterns. Perfect for troubleshooting AI training bottlenecks and optimizing cluster performance.

Backed by Reef Telecom's industry-leading 10-year warranty and technical support. Professional installation services available. This is the network card that powers production AI infrastructure worldwide.

Backed by Reef Telecom's Industry-Leading 10 Year Warranty.
UPC:
PN: MCX623106A-CDAT