Specifications
OPN / Part Number
MCX556A-ECAT
Series
NVIDIA ConnectX-5 VPI (Virtual Protocol Interconnect)
Ports
2 × 100GbE QSFP28 / EDR InfiniBand
Protocols
Ethernet (100/50/40/25/10G) and InfiniBand EDR
Host Interface
PCIe 3.0 x16 (8 GT/s)
RDMA
RoCEv2, InfiniBand verbs
Message Rate
Up to 200 Mpps
MPI Offload
Hardware MPI tag matching and rendezvous
Congestion Control
NVIDIA Congestion Control (DCQCN)
GPUDirect RDMA
Supported (peer-to-peer with NVIDIA GPUs)
Form Factor
PCIe half height, half length
Management
NC-SI, in-band
Compliance
RoHS, CE, FCC Class A
Target Market
HPC, AI clusters, 100G storage fabrics
Overview
ConnectX-5 VPI remains one of the most widely deployed 100G adapters in HPC and AI clusters. The VPI designation means each port can run as 100GbE Ethernet or EDR InfiniBand, selected in firmware — the same card serves both fabrics.
For GPU clusters the relevant features are GPUDirect RDMA (peer-to-peer transfer between GPU memory and the NIC, bypassing host buffers) and hardware MPI tag matching, which together cut message latency and host CPU load on tightly coupled training jobs.
Key Benefits
Dual-protocol VPI: 100GbE or EDR IB per port. GPUDirect RDMA for GPU-direct transfers. 200 Mpps message rate for MPI-heavy codes. 200 Gb/s aggregate in one x16 slot.
Applications
HPC clusters, AI training fabrics, 100G NVMe-oF storage, MPI- and SHMEM-based simulation, and InfiniBand-to-Ethernet gateway nodes.
Request a Quote — NVIDIA CONNECTX-5 MCX556A-ECAT — DUAL 100GBE / EDR VPI ADAPTER
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →