AI-optimized modular spine switch at 460 Tbps (920 Tbps full duplex) supporting up to 576x 800G interfaces for next-generation AI cluster backbone networks.
The Arista 7800R4 Series is a modular AI spine switch platform extending the 7800R3 architecture to native 800G Ethernet and adding the full AI fabric congestion control stack to the spine tier. Available in 4-slot (7804R4) and 8-slot (7808R4) chassis, the platform delivers up to 921.6 Tbps of aggregate switching capacity using 36x QSFP-DD 400G (7800R4-36DD-LC) or 36x OSFP 800G (7800R4-36PD-LC) line cards. Mixing 400G and 800G line cards within the same chassis allows a spine to connect 400G AI fabric leafs on one set of line-card slots while providing 800G inter-spine or uplink connectivity on another.
The 7800R4 brings hardware Priority Flow Control, DCQCN, hardware ECN marking, and RoCE v2 support to the spine chassis form factor, the same congestion-control capabilities present in the 7358X4 and 7368X4 modular AI fabric chassis, enabling lossless Ethernet operation at the spine tier without relying on leaf switches to absorb all congestion. Dual hot-swappable supervisors (7800R4-SUP) maintain control-plane state through failover, and N+1 hot-swappable 3000W AC or DC power supplies deliver the power redundancy expected of a spine-tier component.
The 7800R4-36PD-LC provides 36x OSFP 800G interfaces per slot, enabling 800G spine-to-leaf connectivity for next-generation AI cluster fabrics where GPU nodes attach at 800G. Operating both 400G (7800R4-36DD-LC) and 800G (7800R4-36PD-LC) line cards within the same chassis allows the spine to bridge the 400G-to-800G transition period without chassis replacement.
Hardware PFC and DCQCN at the spine tier allows lossless Ethernet to be enforced end-to-end across both leaf and spine switches, rather than depending on per-hop PFC reactions at only the leaf layer. Spine-level congestion control prevents the backpressure cascade that occurs when leaf-level PFC causes upstream congestion to propagate across the fabric during large AllReduce operations.
At 921.6 Tbps, the 8-slot 7808R4 fully populated with 800G line cards connects more GPU nodes per spine switch than any previous generation of Arista modular spine hardware. This capacity headroom is specifically relevant for the "fat-tree" fabric topology common in large-scale GPU clusters, where spine port count determines the maximum cluster size achievable within a single non-blocking fabric domain.
RoCE v2 (RDMA over Converged Ethernet version 2) support in the spine switching ASIC enables low-latency RDMA path establishment between GPU nodes that attach via different leaf switches, as all inter-leaf GPU-to-GPU traffic traverses the spine. Spine-level RoCE support ensures the RDMA transport operates correctly regardless of which leaf each GPU node connects to, without requiring any application or GPU driver changes.
Full specifications for Arista 7800R4 Series
Download product documentation and resources
Explore other configurations and models that might suit your needs.
Recommended
#DCS-7060PX4-32
Recommended
#DCS-7060DX5-64S
Recommended
#DCS-7060X6-64PE
Our team of experts is ready to help you find the perfect solution for your business needs. Get personalized advice and competitive quotes.
We're here to help with any questions