NVIDIA ConnectX-8 SuperNICs Advance AI Platform Architecture with PCIe Gen6 Connectivity
Table of Contents
- Introduction
- Advancements with ConnectX-8 SuperNICs
- Accelerating Enterprise Workloads with NVIDIA RTX PRO Servers
- GPU-to-NIC Communication
- Conclusion
Introduction
NVIDIA is advancing platform architecture with the introduction of the industry’s first SuperNIC, ConnectX-8, which integrates a PCIe Gen6-capable switch with ultra-high-speed networking in a single device. This innovation aims to streamline server design, reduce component count, and enhance performance scalability for AI workloads.
Advancements with ConnectX-8 SuperNICs
ConnectX-8 addresses the need for PCIe Gen6 connectivity in AI infrastructure by integrating PCIe Gen6 connectivity with an integrated switch, eliminating the need for discrete PCIe switches. This consolidation simplifies board design, reduces component count, and delivers a more cost-effective, scalable architecture for AI infrastructure.
Accelerating Enterprise Workloads with NVIDIA RTX PRO Servers
ConnectX-8 SuperNICs are now integrated with NVIDIA HGX B300 and NVIDIA GB300 NVL72 systems, supporting configurations with up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. The optimized design of NVIDIA RTX PRO Servers with ConnectX-8 SuperNICs enhances network bandwidth per GPU, accelerates collective communication patterns, and improves scalability for multi-GPU and multi-node workloads.
GPU-to-NIC Communication
The optimized architecture of ConnectX-8 provides each GPU with 50 GB/s of bandwidth in a 2:1 GPU-to-NIC configuration, irrespective of PCIe Gen5 or Gen6 support. This innovation enables system builders to scale faster with improved performance and lower total cost of ownership.
Conclusion
By combining a PCIe Gen6 switch and high-performance SuperNIC into a single device, ConnectX-8 simplifies server design, reduces component count, and unlocks high-bandwidth communication paths necessary for modern AI workloads. The result is a more efficient platform with lower total cost of ownership and exceptional performance scalability.