Cisco CAI-INF4-MI210 | MI210 4-Way Infinity Fabric Bridge, 300GB/s Per Link

Q: What is the purpose of the Infinity Fabric Link bridge?
A: The Infinity Fabric Link bridge enables high-speed, direct peer-to-peer communication between multiple AMD Instinct MI210 GPU accelerators within a server, delivering up to 300GB/s bandwidth per link. This dramatically improves performance for HPC, AI training, and large-scale simulations that require fast GPU-to-GPU data exchange.

Q: How many MI210 GPUs can this bridge connect?
A: This 4-way bridge card connects exactly four adjacent AMD Instinct MI210 accelerators in a full mesh topology. All four GPU slots must be fully populated for the bridge to function correctly. For smaller configurations, a 2-way bridge option may be available.

Q: What servers support the CAI-INF4-MI210 bridge?
A: The bridge is designed for Cisco UCS C-series rack servers and similar enterprise platforms that support multiple AMD Instinct MI210 GPU accelerators with PCIe Gen4 slots and sufficient physical spacing for Infinity Fabric bridge installation.

Q: What bandwidth improvement does this bridge provide?
A: Each Infinity Fabric link provides 300GB/s of peer-to-peer bandwidth between GPUs. In a 4-way configuration, the bridge enables up to 1.2TB/s total theoretical GPU bandwidth, significantly outperforming standard PCIe communication for multi-GPU workloads.

Q: Is this bridge compatible with other AMD GPU models?
A: No, this bridge is specifically designed for AMD Instinct MI210 accelerators. Different AMD GPU models such as the MI250X or MI300 series require their own compatible bridge cards and are not interchangeable with the MI210 bridge.