Broadcom BCM957614-N1100G NetXtreme-E 4x25GbE OCP 3.0 NIC

The Broadcom BCM957614-N1100G NetXtreme-E is a high-performance 4-port 25 Gigabit Ethernet network interface card built on Broadcom's advanced NetXtreme-E ASIC architecture. Designed to the Open Compute Project (OCP) 3.0 Mezzanine form factor specification, it integrates directly into compatible OCP 3.0 server slots, eliminating the need for a dedicated PCIe slot and reducing cabling overhead in modern hyperscale and enterprise rack deployments. The adapter delivers an aggregate bandwidth of 100GbE across four independent 25GbE SFP28 ports, supporting both direct-attach copper (DAC) and active optical cable (AOC) connectivity for flexible datacenter fabric integration.

The BCM957614-N1100G incorporates Broadcom's RoCE (RDMA over Converged Ethernet) v1 and v2 support, enabling ultra-low-latency, CPU-bypass data transfers critical for distributed storage, high-performance computing, and AI/ML training workloads. The adapter supports advanced offload capabilities including stateless TCP/UDP/IP checksum offload, Large Send Offload (LSO), Large Receive Offload (LRO), and VXLAN/GENEVE tunnel offload, substantially reducing host CPU overhead in virtualized and containerized environments. SR-IOV support allows the card to present multiple virtual functions (VFs) to hypervisors and container orchestration platforms, enabling granular network resource allocation across virtual machines and Kubernetes pods.

Tailored for enterprise datacenter, cloud infrastructure, and AI workload environments, the NetXtreme-E BCM957614-N1100G is fully compatible with mainstream operating systems including Linux and Windows Server, and integrates with leading virtualization stacks such as VMware vSphere and Microsoft Hyper-V. Its OCP 3.0 form factor makes it an ideal choice for OCP-compliant server platforms from major brands targeting dense, power-efficient rack deployments in hyperscale, colocation, and enterprise AI infrastructure.

Ideal for

• AI and machine learning cluster networking requiring high-bandwidth, low-latency RDMA (RoCE v2) interconnects between GPU servers for distributed training workloads
• High-density virtualized datacenter environments leveraging SR-IOV to allocate dedicated 25GbE virtual functions across large numbers of VMs without PCIe slot consumption
• NVMe-over-Fabrics (NVMe-oF) storage networking connecting all-flash arrays to compute nodes over a 25GbE RoCE fabric with minimal CPU overhead
• Containerized and Kubernetes-based cloud-native workloads requiring hardware-accelerated VXLAN/GENEVE overlay network offload to reduce hypervisor CPU cycles
• OCP 3.0 compliant hyperscale and colocation server deployments where mezzanine form factor integration maximizes rack density and reduces cabling complexity
• High-performance computing (HPC) environments needing multi-port 25GbE connectivity with RDMA capabilities to support low-latency MPI communication between compute nodes

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.