NVIDIA Mellanox® MCB194A-FCAT — Connect-IB™ Dual-Port FDR 56Gb/s InfiniBand Host Channel Adapter, PCIe 3.0 x16
Technical Identity
The NVIDIA Mellanox MCB194A-FCAT is a Connect-IB™ generation, dual-port FDR 56 Gb/s InfiniBand Host Channel Adapter (HCA) on a PCIe 3.0 x16 host interface, with two QSFP ports. Built on Mellanox's Connect-IB silicon — the HCA generation that sits between ConnectX-3 and ConnectX-4 and the first card to introduce Dynamically Connected Transport (DCT) — it is designed for HPC clusters, low-latency financial trading fabrics, and high-throughput storage back-ends. This is an InfiniBand-only adapter; it does not support Ethernet and is not a VPI card.
Specifications
What This Is NOT
NOT a ConnectX-3, ConnectX-4, ConnectX-5, ConnectX-6, or ConnectX-7 card. Connect-IB is its own ASIC family with separate firmware (PSID) and driver feature gates.
NOT an EDR (100 Gb/s), HDR (200 Gb/s), or NDR (400 Gb/s) HCA. Maximum link rate is FDR 56 Gb/s per port.
NOT VPI / NOT Ethernet-capable. The Connect-IB ASIC does not implement Ethernet — for dual-protocol use a ConnectX-3 VPI or ConnectX-4/5/6 VPI card.
NOT a SmartNIC or DPU. No on-card ARM cores or programmable data path — see BlueField-2 / BlueField-3 for DPU functionality.
NOT a switch. This is a host adapter and requires an FDR-capable InfiniBand switch upstream.
NOT bundled with cables, transceivers, or brackets unless explicitly stated.
Compatibility
InfiniBand Switches (pairs with): Mellanox SX6018, SX6025, SX6036, SX6512, SX6710, SX6790 FDR InfiniBand switches; also backward compatible with QDR switches (IS5030, IS5035, SX6036 in QDR mode).
Cables & Transceivers (pairs with): Mellanox MC2207130 / MC2207128 / MC220731V series QSFP+ FDR copper (DAC), MC2207310 / MC220731V series active optical cables (AOC), and Mellanox LinkX FDR optical transceivers + OM3/OM4 multi-mode fibre.
Servers: Any server with a free PCIe Gen 3.0 x16 slot — Dell PowerEdge R720/R730, HPE ProLiant DL380 Gen8/Gen9, Supermicro X9/X10, Lenovo ThinkSystem (validated under MLNX_OFED). Will operate in a PCIe 3.0 x8 slot at reduced aggregate bandwidth.
Drivers: MLNX_OFED 2.x through 4.x (Connect-IB is end-of-active-development in MLNX_OFED 5.x+; for newer driver stacks migrate to ConnectX-4 or later).
GPUDirect RDMA: Validated with NVIDIA Kepler and Pascal Tesla GPUs (K20, K40, K80, P100) on CUDA 6.5–9.x.
Operating Systems: RHEL/CentOS 6.x–7.x, SLES 11/12, Ubuntu 14.04/16.04/18.04, Windows Server 2012 R2 / 2016, VMware ESXi 5.x/6.x.
FAQ
What is the NVIDIA Mellanox MCB194A-FCAT?
A: The MCB194A-FCAT is a Connect-IB™ generation, dual-port FDR 56 Gb/s InfiniBand Host Channel Adapter on a PCIe 3.0 x16 host interface, with 2× QSFP ports. It is InfiniBand-only and does not support Ethernet.
Is the MCB194A-FCAT a ConnectX adapter?
A: MCB194A-FCAT: No. Connect-IB is its own ASIC family — released between ConnectX-3 and ConnectX-4 — with separate firmware images and driver paths. It is not interchangeable with ConnectX-3 or ConnectX-4 cards.
Can the MCB194A-FCAT run Ethernet?
A: MCB194A-FCAT: No. The Connect-IB silicon is InfiniBand-only. For dual-protocol (VPI) cards that switch between InfiniBand and Ethernet, use ConnectX-3 VPI, ConnectX-4 VPI, ConnectX-5 VPI, or ConnectX-6 VPI.
Which switches and cables does it work with?
A: MCB194A-FCAT: Any FDR InfiniBand switch — Mellanox SX6xxx series — using standard QSFP+ FDR DAC or AOC cables (Mellanox MC22071xx / MC22073xx families) and LinkX FDR optical transceivers. It will also negotiate down to QDR (40 Gb/s) on legacy QDR fabrics.
What PCIe slot does it require?
A: MCB194A-FCAT: PCI Express Gen 3.0 x16 for full dual-port FDR bandwidth. It will fit and operate in a PCIe 3.0 x8 slot, but aggregate throughput is approximately halved.
Is this a new or refurbished unit?
A: MCB194A-FCAT: New, factory-sealed in original NVIDIA Mellanox packaging where included.
Why T.E.S IT-SOLUTIONS
Every MCB194A-FCAT shipped by T.E.S IT-SOLUTIONS is checked (see Inspection and checks). Anti-static professional packaging with global B2B express shipping, multi-currency invoicing, and direct technical pre-sales consultation for HPC, GPU-cluster, and InfiniBand-storage deployments — speak to our engineering team, not a call-center. See tes-itsolutions.com.