- AMD SHIPS POLLARA 400 AI NIC FOR HIGH SPEED AU NETWORKS
- Admits the Ultra Ethernet standard with RDMA and RCCL for efficient communication
- Future Vulcano 800g AI Nic Targets Pcie Gen6 and GPU Groups at Rack Scale
AMD has begun to send the network card thinking 400 AI, part of the company’s impulse by the open data of high speed data.
Designed for PCIE GEN5 systems, the card admits the Ultra Ethernet consortium standard (EUC), whose objective is to transform Ethernet for AI and HPC at scale.
The card offers RDMA support and is optimized for collective scale communication using RCCL, AMD’s alternative to NCCL.
Vulcano 800g Ai nic for a 2026 launch
AMD says that Pollara offers about 10% better RDMA performance than NVIDIA Connectx-7 and approximately 20% better than Broadcom Thor2. In GPU groups, these profits help reduce inactivity time and improve the efficiency of the workload.
NIC uses a personalized processor with flexible transport protocols, loading balance and switching by error. It can redirect traffic during congestion and maintain GPU connectivity during failures.
The card presents a medium height, half length design and admits PCIe Gen5 X16, which offers multiple port configurations that include 1x400g, 2x200g and 4x100g. It admits up to 400 Gbps of bandwidth and integrates monitoring tools to improve observability and reliability at the cluster level.
AMD states that performance increases up to 6x in large -scale implementations, especially when it is reduced to hundreds of thousands of processors.
For individual workloads, the company reports an AI work performance up to 15% faster and up to 10% improved network reliability through features such as rapid error switching, selective retransmission and congestion management.
With the EUC 1.0 specification now finished, the company is aimed at Hipperscalers. Oracle Cloud will be among the first to adopt technology.
Looking at 2026, AMD says it is intended to launch the Vulcano 800g Ai nic for PCIE Gen6 systems (Pollara and Vulcano are the names of two volcanoes in Italy).
That NIC will admit both Ultra Ethernet and Uualink to enable the networks of scale and expansion for large workloads of AI. Vulcano is part of the AMD Helios Rack Scale architecture, scheduled for 2026.
AMD is positioning Vulcano as an open alternative of multiple suppliers to Nvidia’s Connectx-8. Its success may depend on the speed with which the broader ecosystem can adapt and support the new network standards.
Writing about the two network cards, Patrick Kennedy in Servethehome Look “, at the end of the day, if you want to play in 2026 clusters of AI, it does not need only AI chips, but also the ability to climb and climb. AMD may sound a lot like Nvidia’s play book because it is necessary. On the other hand, supporting open standards is very different from what Nvidia is doing when leaning in multiple interviews and open standards.” “
