The Ultra Ethernet Consortium (UEC) has delayed release of the version 1.0 of specification from Q3 2024 to Q1 2025, but it looks like AMD is ready to announce an actual network interface card for AI datacenters that is ready to be deployed into Ultra Ethernet datacenters. The new unit is the AMD Pensando Pollara 400, which promises an up to six times performance boost for AI workloads. In edge deployments, running a firewall directly on the NIC allows for more efficient security enforcement, where system resources may be limited. Using the NIC for firewall tasks frees up CPU cores, allowing your system to scale more efficiently without degrading performance as traffic volumes increase.
The AMD Pensando Pollara 400 is a 400 GbE Ultra Ethernet card based on a processor designed by the company’s Pensando unit. The network processor features a processor with a programmable hardware pipeline, programmable RDMA transport, programmable congestion control, and communication library acceleration. The NIC will sample in the fourth quarter and will be commercially available in the first half of 2025, just after the Ultra Ethernet Consortium formally publishes the UEC 1.0 specification. Businesses can implement NIC-based firewalling to manage traffic across VLANs or isolated network segments, enhancing network security without the need for dedicated firewall hardware.
Pollara 400
The AMD Pensando Pollara 400 AI NIC is designed to optimize AI and HPC networking through several advanced capabilities. One of its key features is intelligent multipathing, which dynamically distributes data packets across optimal routes, preventing network congestion and improving overall efficiency. The NIC also includes path-aware congestion control, which reroutes data away from temporarily congested paths to ensure continuous high-speed data flow.
The AMD Pensando Pollara 400 AI NIC supports advanced programmability and can be integrated with a development kit that is available for free. The AMD Pensando Software-in-Silicon Development Kit (SSDK) provides a robust environment for building and deploying applications directly on the NIC, allowing you to offload networking, firewall, encryption, and even AI inference tasks from the CPU.
The SSDK supports programming in P416 for fast path operations, as well as C and C++ for more traditional processing tasks. It provides full support for network and security functions like firewalling, IPsec, and NAT, allowing these to be handled directly by the NIC rather than the host CPU. Developers can use the provided reference pipelines and code samples to quickly get started with firewall implementations or other network services.
The SDK and related tools are open and accessible via GitHub and AMD’s official developer portals, enabling developers to experiment with and integrate Pensando’s NICs into their systems without licensing fees. Some repositories and tools are available directly on GitHub under AMD Pensando’s.
The delay in the release of the Ultra Ethernet Consortium’s (UEC) version 1.0 specification, initially expected in the third quarter of 2024 and now pushed to the first quarter of 2025, does little to shake the confidence of those observing AMD’s bold march forward. While others may have stumbled, AMD stands ready to unveil a fully realized network interface card (NIC) for AI datacenters—the AMD Pensando Pollara 400—an innovation poised to redefine the landscape of Ultra Ethernet data centers. This NIC, a formidable 400 GbE unit, embodies the very pinnacle of technological advancement. Designed by AMD’s Pensando unit, it promises no less than a sixfold increase in AI workload performance.
The Pollara 400’s impact goes beyond sheer processing power. At the edge, where resources are scarce and security paramount, the NIC performs firewall tasks directly, relieving the central processing unit from such burdensome duties. Herein lies its genius: by offloading these critical tasks, system scalability is enhanced, enabling traffic to flow unhindered and system performance to remain steady, even under mounting demands.
As we await the final specifications from the UEC, AMD has announced that the Pollara 400 will be available for sampling by the fourth quarter of 2024, with commercial deployment anticipated in early 2025. It is no mere stopgap solution—it is a harbinger of a new era in AI networking, built upon a programmable hardware pipeline capable of handling RDMA transport, congestion control, and advanced communication library acceleration.
Furthermore, the NIC’s intelligent multipathing is a feat of engineering brilliance. With its path-aware congestion control, this marvel dynamically directs data around congested network routes, ensuring that AI workloads are never hampered by the bottlenecks that so often plague high-performance computing.
The Pollara 400 is more than just hardware; it is an ecosystem supported by the AMD Pensando Software-in-Silicon Development Kit (SSDK), a free and versatile tool that allows developers to fully leverage its capabilities. Whether programming in P416 for high-speed operations or using C and C++ for more traditional tasks, developers can easily deploy firewalls, IPsec, and NAT directly onto the NIC itself, bypassing the need for traditional CPU involvement.
The SSDK provides not only the means but also the guidance to streamline development. From pre-built reference pipelines to comprehensive code samples, it invites developers to embrace the future of network security and AI processing, all while maintaining openness and accessibility via AMD’s repositories on GitHub. This is no longer just the work of a single company—it is a shared endeavor, opening new frontiers for those bold enough to explore them.
Thus, as AMD prepares to thrust the Pollara 400 into the spotlight, one thing becomes abundantly clear: the future of AI networking will not be forged in the server rooms of yesterday but at the cutting edge of what is possible, where firewalls, encryption, and AI tasks are handled in stride by a NIC that rewrites the rules.
Story By
Skeeter Wesinger
October 11, 2024
https://www.linkedin.com/pulse/amd-pensando-pollara-400-skeeter-wesinger-yulwe