Cisco UCSC-P-APS400GF | Pollara 400 AI Network Adapter, 400GbE QSFP112, PCIe 5.0 x16

Q: What is the AMD Pensando Pollara 400 AI NIC designed for?
A: The Pollara 400 AI NIC is specifically designed for high-performance AI data center environments, optimized for GPU-to-GPU communication in scale-out AI clusters. It excels in AI/ML workloads, generative AI applications, and large language model training where ultra-low latency and high throughput are critical.

Q: What makes this NIC different from traditional network adapters?
A: The Pollara 400 features full hardware programmability through AMD's third-generation Pensando P4 engine, enabling customizable transport protocols and intelligent packet processing. It supports Ultra Ethernet Consortium standards with advanced features like adaptive packet spray, path-aware congestion control, selective retransmission, and out-of-order packet handling to optimize bandwidth utilization and reduce GPU idle time.

Q: Which RDMA protocols does the Pollara 400 support?
A: The adapter supports multiple RDMA protocols including RoCEv2 (RDMA over Converged Ethernet version 2) and UEC RDMA, ensuring compatibility with various AI workloads and network fabrics. It provides vendor-agnostic compatibility for flexible AI infrastructure deployment.

Q: What PCIe generation and lane configuration is required?
A: The Pollara 400 requires a PCIe Gen 5.0 x16 slot to achieve full 400 Gbps performance with 32 GT/s interconnect bit rate. The HHHL form factor makes it compatible with standard server chassis and is commonly deployed in 1:1 ratio with GPUs in AI server configurations.

Q: Can this NIC support multiple Ethernet speeds?
A: Yes, the Pollara 400 is versatile and supports multiple Ethernet speeds including 25GbE, 50GbE, 100GbE, 200GbE, and 400GbE, providing flexibility for various network configurations and migration paths within data center environments.