Arista 7368X4 Series

The Arista 7368X4 Series is a modular 400G AI fabric switch designed for large-scale GPU training cluster interconnects where deterministic, lossless Ethernet operation is required at all traffic levels. Available in 4-slot (7368X4-4) and 8-slot (7368X4-8) chassis configurations with the 7368X4-LC-36DD 36x QSFP-DD 400G line card, the platform reaches 230.4 Tbps maximum aggregate capacity. Like its larger sibling the 7358X4, the 7368X4 supports Priority Flow Control, DCQCN, and hardware ECN marking as the congestion management foundation for RoCE v2 workloads running NCCL AllReduce operations across thousands of GPU nodes.

The 7368X4 occupies a mid-range position in the Arista AI fabric switch family: smaller than the 7358X4 in total slot count and maximum capacity, but sharing the same congestion control hardware architecture and the same EOS software feature set. This consistency allows operators to run identical fabric configurations across both chassis sizes, critical for organisations building multi-pod AI cluster facilities where the pod-level fabric (7368X4) and the inter-pod spine (7358X4 or 7800R4) need to operate as a unified, coherently managed forwarding domain.