- AMD Ships Pollara 400 AI Nic for Open High Speed IA networking
- Supports the ultra ethernet standard with RDMA and RCCL for effective communication
- Future Vulcano 800g AI Nic Target Pcie Gen6 and GPU clusters on a scale at rack
AMD began to ship the Pensando Pollara 400 AI network card, part of the thrust of the company’s high -speed data center.
Designed for PCIe Gen5 systems, the card supports the Ultra Ethernet Consortium (UEC) standard, which aims to transform Ethernet for large -scale AI and HPC.
The card offers RDMA support and is optimized for collective communication on a scale using the RCCL alternative, AMD in NCCL.
Vulcano 800g ai nic targeting a launch in 2026
AMD says Polara offers around 10% of better RDMA performance than Connectx-7 from Nvidia and around 20% better than Broadcom’s Thor2. In the heavy clusters of the GPU, these gains help reduce the time of inactivity and improve the efficiency of the workload.
The NIC uses a personalized processor with the support of flexible transport protocols, load balancing and tilting routing. It can relaunch traffic during congestion and maintain the connectivity of the GPU during failures.
The card has a half-height design and half a lecturer and supports PCIe Gen5 X16, offering several port configurations, including 1x400g, 2x200g and 4x100g. It supports up to 400 gbit / s of bandwidth and incorporates monitoring tools to improve observability and reliability in the cluster.
AMD claims that performance is increasing up to 6x in large -scale deployments, especially when put to scale at hundreds of thousands of processors.
For individual workloads, the company reports up to 15% professional performance on the AI and up to 10% of the reliability of the improved network thanks to functionalities such as rapid tilting, selective retransmission and congestion management.
With the UEC 1.0 specification now finalized, the company targets hyperscalers. Oracle Cloud will be among the first to adopt technology.
Looking at 2026, AMD says that it intends to launch the Pensando Vulcano 800g AI Nic for PCIe Gen6 systems (Pollara and Vulcano are the names of two volcanoes in Italy).
This NIC will support both Ultra Ethernet and Ualink to allow a scaling and scaling of the network for major AI workloads. Vulcano is part of the architecture on the Rack of Helios d’AMD, fixed for 2026.
AMD positions Vulcano as an open and multi-sales alternative to Connectx-8 from Nvidia. Its success may depend on the speed with which the wider ecosystem can adapt and support new networking standards.
Writing on the two networking cards, Patrick Kennedy in Serve Observe: “At the end of the day, if you want to play in the clusters of IA 2026, you not only need a AI chips, but also the possibility of developing and getting to the scale. AMD having a network card may look much like Nvidia is necessary.
