AMD Launched High-Speed AI Network Interface Cards
The new 400 and 800 Gbps network interfaces aim to solve congestion management bottlenecks in large AI GPU clusters.
Updated on Oct. 13, 2026 in Semiconductors
Live Poll
Is now a good time for your organization to upgrade network infrastructure for AI workloads?
AMD has announced the Pensando Pollara 400 and Vulcano 800 network interface cards (NICs), hardware designed to improve data throughput for AI workloads. The Pollara 400 is now validated for use with AMD Instinct MI350 and MI355 GPU clusters.
Why it matters
Conventional Ethernet transports often struggle with the congestion management and multipath requirements of large-scale AI training. These new NICs address those limitations to enable more efficient cluster scaling.
The Pollara 400 supports 400 Gbps throughput, while the Vulcano 800 scales to 800 Gbps. Using a three-NIC configuration per GPU, the system achieves 2.4 Tbps of total bandwidth.
The players
AMD
A semiconductor company that designs high-performance CPUs, GPUs, and networking hardware for data centers and AI clusters.
The details
The architecture utilizes a hardware pipeline based on P4, a programming language for network devices, to manage high-speed traffic. The system implements Multipath Reliable Connection to perform packet-spray load balancing, which distributes data across multiple network paths to reroute traffic away from congested links. Additional features include selective retransmission and rapid fault detection to improve reliability across large GPU clusters.
Timeline
October 13, 2026: AMD launched the Pollara 400 and Vulcano 800 networking hardware.
The Tech Race
AMD is positioning this hardware to meet the evolving standards set by the Ultra Ethernet Consortium for high-performance AI data transfer. The effort follows a broader industry trend of moving congestion control logic from software to the network edge.
These cards are intended for enterprise data centers managing large-scale AI training workloads using Instinct GPUs. Users can expect the hardware to enable higher throughput in clusters that utilize RoCEv2 and the emerging Ultra Ethernet protocols.
The takeaway
AMD aims to optimize AI cluster efficiency by offloading complex networking logic to specialized hardware. Watch for the forthcoming validation of the Vulcano 800 platform to see if the company can sustain these throughput gains at the 800 Gbps tier.
Further reading
For more on how new hardware is evolving to support AI scale, see our coverage in Semiconductors.
Live Poll
Is now a good time for your organization to upgrade network infrastructure for AI workloads?
