Cisco CAI-NVL2-H100 | NVLink 2-Way Bridge for H100 NVL GPU Interconnect

Q: Is this bridge compatible with standard H100 PCIe or H100 SXM GPUs?
A: This bridge is specifically designed for H100 NVL GPUs only. Standard H100 PCIe cards use a different NVLink bridge configuration, and H100 SXM GPUs use integrated NVLink 4.0 fabric that does not require external bridges.

Q: How many bridges do I need to connect two H100 NVL cards?
A: You need three NVLink bridges per pair of adjacent H100 NVL cards to achieve full 600 GB/s bidirectional bandwidth. Each bridge spans two PCIe slots and all three must be installed for the connection to function at peak performance.

Q: What performance improvement does NVLink provide over PCIe?
A: NVLink delivers 600 GB/s bidirectional bandwidth between H100 NVL GPUs, approximately 10x the bandwidth of PCIe Gen4. This dramatically reduces data transfer latency for large model training and inference workloads that require frequent GPU-to-GPU communication.

Q: Do I need special drivers or software to use NVLink?
A: Yes, you need NVIDIA drivers that support NVLink connectivity. Ensure your AI frameworks and CUDA applications are configured to recognize and utilize the NVLink interconnect for multi-GPU operations. Most modern ML frameworks automatically detect NVLink when present.

Q: Can I use this bridge with other GPU models or brands?
A: No, this bridge is exclusively compatible with NVIDIA H100 NVL GPUs. It cannot be used with A100, H200, or any other GPU generation or model due to physical connector and electrical protocol differences.