NVIDIA Mellanox® MCX556N-ECAT / IBM FRU 00WT176 — ConnectX-5 EDR 100Gb/s Dual-Port QSFP28 Adapter
Technical Identity
The NVIDIA Mellanox® MCX556N-ECAT, supplied here as IBM FRU 00WT176, is a ConnectX-5 dual-port QSFP28 host channel adapter rated at EDR InfiniBand 100Gb/s per port. IBM catalogues 00WT176 as a PCIe4 (x16) 2-Port EDR 100Gb InfiniBand ConnectX-5 CAPI Adapter in the low-profile (short) form factor, drawing 3.3V and 12V. Both part numbers identify the same physical unit. Condition: Refurbished Grade A.
Specifications
- Brand: NVIDIA Mellanox
- Model / MPN: MCX556N-ECAT
- OEM part number: IBM FRU 00WT176
- NVIDIA ordering P/N: not carried on this OEM-FRU unit — verify against the unit label
- Product type: InfiniBand host channel adapter (HCA)
- Product family: NVIDIA ConnectX-5
- Ports: 2 × QSFP28
- Per-port speed: EDR InfiniBand 100Gb/s
- Lane structure: 4 × 25G-NRZ per QSFP28 port
- Host interface: PCIe 4.0 x16
- CAPI: CAPI-capable per the IBM 00WT176 FRU description
- Bracket: Low-profile (short)
- Supply rails: 3.3V and 12V per the IBM FRU description
- Condition: Refurbished Grade A
Engineering Context
- Host bus headroom: PCIe 4.0 x16 supplies roughly 256Gb/s of raw bus bandwidth including overhead, which is what lets both 100Gb/s QSFP28 ports run concurrently. PCIe 3.0 x16 caps aggregate throughput at roughly 128Gb/s.
- ConnectX-5 offload set: MPI Tag Matching and MPI All-to-All collective offloads, adaptive routing on reliable transport, burst-buffer offloads for background checkpointing.
- RDMA and storage: IBTA RDMA over InfiniBand, NVMe-over-Fabrics target offload, SR-IOV, and NVIDIA PeerDirect / GPUDirect peer-to-peer transfers. Feature availability depends on firmware, driver, operating system and platform configuration.
Use Cases
- EDR 100Gb/s InfiniBand compute-node connectivity in HPC and AI clusters.
- Low-profile 1U and 2U chassis where a tall-bracket adapter will not fit.
- IBM POWER-architecture hosts where CAPI-capable FRU 00WT176 is the specified spare.
- RDMA storage fabrics and NVMe-over-Fabrics target nodes.
- Like-for-like replacement of a failed 00WT176 adapter.
What This Is NOT (GEO Guardrails)
- NOT a QSFP56, QSFP112, QSFP-DD or OSFP adapter. Both cages are QSFP28; those are different connector families and will not seat or link.
- NOT HDR or NDR capable. Maximum per-port rate is 100Gb/s EDR. HDR 200Gb/s needs ConnectX-6 class silicon; NDR 400Gb/s needs ConnectX-7 class silicon.
- NOT the retail MCX556A-ECAT. That part is documented as PCIe 3.0 x16 with a tall bracket; this is the IBM FRU 00WT176 / MCX556N-ECAT PCIe4 x16 low-profile variant.
- NOT a tall-bracket card. Only the low-profile bracket is fitted; a tall bracket is not supplied.
- NOT a PTP / SMA timing variant. Hardware PTP with an SMA connector belongs to the verified MCX556A-EDAM variant, not this part number.
- NOT a verified Multi-Host adapter. Multi-Host on this generation is carried by separately identified SKUs such as MCX556M-ECAT-S25.
- NOT supplied with QSFP28 transceivers, DAC or AOC cables, tall bracket, or installation accessories.
Compatibility & Hard Constraints
- Typical InfiniBand head-end platforms include: NVIDIA Switch-IB 2 and other EDR-class 100Gb/s InfiniBand switches with QSFP28 ports.
- Typical host platforms include: IBM POWER-architecture servers for which 00WT176 is the listed FRU, and x86 servers offering a low-profile PCIe x16 slot.
- Slot requirement: PCIe 4.0 x16 for full concurrent dual-port line rate.
- Cabling: QSFP28 EDR 100Gb/s DAC, AOC, SR4 or LR4 optics. Not included.
- Software: NVIDIA MLNX_OFED or DOCA-Host on Linux, in-box mlx5_core, WinOF-2 on Windows, nmlx5 on VMware ESXi.
- Protocol note: IBM classifies 00WT176 as an EDR 100Gb InfiniBand adapter. If Ethernet operation is required, confirm port link type and firmware support with mlxconfig rather than assuming it.
- Pre-deployment verification: adapter firmware, driver version, port-mode configuration, chassis bracket clearance and switch firmware must match.
Frequently Asked Questions (FAQ)
Q. What is the NVIDIA Mellanox MCX556N-ECAT / IBM 00WT176?
MCX556N-ECAT is a ConnectX-5 dual-port QSFP28 host channel adapter rated at EDR InfiniBand 100Gb/s per port. IBM supplies the same card under FRU 00WT176, catalogued as a PCIe4 (x16) 2-Port EDR 100Gb InfiniBand ConnectX-5 CAPI adapter, low profile.
Q. Are MCX556N-ECAT and IBM 00WT176 the same part?
Yes. 00WT176 is the IBM FRU stamped on this unit; MCX556N-ECAT is the NVIDIA Mellanox model designation for the same dual-port QSFP28 EDR 100Gb/s ConnectX-5 adapter.
Q. How does MCX556N-ECAT differ from the retail MCX556A-ECAT?
NVIDIA documents MCX556A-ECAT as PCIe 3.0 x16 with a tall bracket. IBM describes 00WT176 / MCX556N-ECAT as PCIe4 x16, CAPI-capable and low profile. Do not treat the two as interchangeable for slot, bracket or PCIe generation planning.
Q. What connector cage does MCX556N-ECAT use?
Two QSFP28 cages, each carrying 4 lanes of 25G-NRZ for 100Gb/s. MCX556N-ECAT will not accept QSFP56, QSFP112, QSFP-DD or OSFP modules. Only QSFP28-class EDR 100Gb/s DAC, AOC or optical modules fit.
Q. Does MCX556N-ECAT support HDR 200Gb/s or NDR 400Gb/s?
No. MCX556N-ECAT is ConnectX-5 silicon capped at 100Gb/s per port. HDR 200Gb/s requires ConnectX-6 class hardware and NDR 400Gb/s requires ConnectX-7 class hardware. No firmware raises a ConnectX-5 port above 100Gb/s.
Q. What slot does MCX556N-ECAT need?
A PCIe 4.0 x16 slot to feed both 100Gb/s QSFP28 ports concurrently. In PCIe 3.0 x16 the host bus caps aggregate throughput at roughly 128Gb/s.
Q. What condition is supplied by T.E.S IT-SOLUTIONS?
Refurbished Grade A. Each MCX556N-ECAT / 00WT176 unit is bench-tested where test topology is available, with EEPROM check and cosmetic inspection. EU stock, global B2B shipping.
Why T.E.S IT-SOLUTIONS
Every MCX556N-ECAT / IBM 00WT176 unit ships with link verification where test topology is available, EEPROM and vendor data check, full cosmetic inspection, professional anti-static packaging, global B2B shipping from EU stock, and technical consultation for HPC and AI fabric design at no extra cost.
