NVIDIA Mellanox MBF1L516A-ESCAT BlueField Data Processing Unit (DPU), VPI Dual-Port 100GbE / InfiniBand EDR, PCIe Gen 4.0: This is a programmable BlueField SmartNIC / DPU adapter card with two QSFP28 ports operating as Virtual Protocol Interconnect (VPI). Each port can be configured for either 100GbE Ethernet or InfiniBand EDR (100Gb/s). The host interface is PCIe Gen 4.0. This is a BlueField DPU — it integrates ARM compute, on-card memory, and the ConnectX-class network ASIC; it is not a plain ConnectX NIC, not a switch, and not a transceiver. This is a refurbished unit.
Engineering Context
BlueField DPUs combine a ConnectX-derived network interface with on-board ARM cores running a full Linux distribution (NVIDIA DOCA / BFB image). The platform is designed to offload software-defined networking, storage, and security functions from the host CPU and execute them on the DPU itself — including OVS data-plane acceleration, NVMe-over-Fabrics target/initiator emulation, IPsec/TLS bulk-encryption offload, and isolated infrastructure control planes. VPI mode means each port can be flipped between Ethernet and InfiniBand at firmware level without swapping hardware. PCIe Gen 4.0 doubles the per-lane host bandwidth of Gen 3, which matters for sustained dual-port 100GbE/EDR throughput.
Deployment & Use Cases
Bare-metal cloud provider tenancy where the DPU runs the hypervisor management plane in isolation from the tenant CPU. Software-defined storage targets running NVMe-over-Fabrics with on-DPU NVMe emulation. Inline IPsec/TLS termination at the network edge of a host. East-west traffic firewalling inside hyperconverged clusters. Mixed Ethernet/InfiniBand fabrics where a single adapter must serve both protocols across reboots.
Technical Specifications
OEM: NVIDIA Mellanox
Part Number: MBF1L516A-ESCAT
Family: BlueField DPU / SmartNIC
Ports: 2 × QSFP28
Port Modes (VPI): 100GbE Ethernet or InfiniBand EDR (100Gb/s)
Host Interface: PCIe Gen 4.0
On-Card Compute: ARM cores running NVIDIA BFB / DOCA software stack
RDMA: RoCE v1, RoCE v2 (Ethernet); native verbs (InfiniBand)
Virtualization: SR-IOV with PCIe sub-functions
Form Factor: PCIe HHHL adapter card with bracket
Software: NVIDIA DOCA SDK, BlueField BFB OS
Condition: Refurbished — Tested by T.E.S IT-SOLUTIONS
Compatibility & Hard Constraints
Requires a host with a PCIe Gen 4.0 x16 slot for full bandwidth; Gen 3.0 slots will physically fit but cap aggregate throughput. The DPU operating system (BFB) must match the firmware revision of the card — confirm version compatibility before flashing. Driver support requires NVIDIA DOCA / OFED on the host plus a BFB image on the DPU itself. VPI mode switching between Ethernet and InfiniBand requires re-flashing the port mode and a card reset; per-port mode is set in firmware. Not interchangeable with a plain ConnectX NIC of the same generation — application offload images must be built for the BlueField DPU. Not a switch — does not provide multi-host bridging on its own.
Why Choose T.E.S IT-SOLUTIONS
Every BlueField DPU is verified for boot, link training on both QSFP28 ports, PCIe Gen 4 link width, and BFB OS bring-up on the embedded ARM cores before shipment. T.E.S IT-SOLUTIONS provides expert consultation on DPU offload architectures, global shipping, and 1–3 day order handling.