1-832-478-9159 Monday - Friday 9 am - 5 pm
ConnectX-6

ConnectX-6 NICs

Top of page ↑
ConnectX-7

ConnectX-7 NICs

Top of page ↑
ConnectX-8

ConnectX-8 NICs

Top of page ↑
NVIDIA ConnectX-8 NIC
NVIDIA Networking

ConnectX-8 Network Adapters

High-performance 400GbE Ethernet adapters designed for AI infrastructure, HPC clusters, and hyperscale cloud data centers. ConnectX-8 delivers next-generation throughput with ultra-low latency, advanced RDMA (RoCEv2) acceleration, and PCIe Gen6 performance for large-scale distributed AI training and inference workloads.

Engineered for modern AI factories and high-density GPU environments, ConnectX-8 provides the optimal balance of bandwidth, efficiency, and scalability for enterprise and hyperscale networking.

400GbE Max Speed
PCIe Gen6 Interface
QSFP112 Connectivity
RDMA Acceleration
NVIDIA ConnectX-7 NIC
NVIDIA Networking

ConnectX-7 Network Adapters

Next-generation high-performance Ethernet adapters delivering up to 400GbE bandwidth for AI, HPC, and cloud-scale infrastructure.

Built on PCIe Gen5 architecture, ConnectX-7 enables extreme throughput, ultra-low latency, and advanced RDMA (RoCEv2) acceleration for data-intensive workloads such as AI training clusters, high-frequency trading, and large-scale distributed storage. With support for 400GbE speeds and modern high-speed connectivity, ConnectX-7 is engineered for next-level scalability in modern data centers.

400GbE Max Speed
PCIe Gen5 Interface
QSFP112 Connectivity
RDMA Acceleration
NVIDIA ConnectX-6 NIC
NVIDIA Networking

ConnectX-6 Network Adapters

High-performance 200GbE and 100GbE Ethernet adapters designed for AI, cloud, and high-performance computing workloads. ConnectX-6 delivers industry-leading throughput, low latency, and hardware-accelerated RDMA (RoCEv2) to support modern AI training, distributed computing, and data center storage environments.

Built on PCIe Gen4 architecture, ConnectX-6 provides an optimal balance of performance and efficiency for scalable AI infrastructure, hyperscale cloud deployments, and enterprise networking.

200GbE Max Speed
PCIe Gen4 Interface
QSFP56 Connectivity
RDMA Acceleration
NVIDIA CONNECTX

ConnectX Generations

Compare the networking capabilities, host interfaces, port configurations, and supported speeds across NVIDIA ConnectX-6, ConnectX-7, and ConnectX-8 adapters.

Specification
ConnectX-6
Ethernet Adapter
ConnectX-7
Ethernet / InfiniBand
ConnectX-8
SuperNIC
Maximum Ethernet Speed 200 Gb/s 400 Gb/s 400 Gb/s
Maximum InfiniBand Speed 200 Gb/s HDR 400 Gb/s NDR 800 Gb/s XDR
PCI Express PCIe Gen 4 x16 PCIe Gen 5 x16 PCIe Gen 6 x16
Ethernet Speeds 1 / 10 / 25 / 40 / 50 / 100 / 200 Gb/s 10 / 25 / 40 / 50 / 100 / 200 / 400 Gb/s 10 / 25 / 40 / 50 / 100 / 200 / 400 Gb/s
Primary Network Connector QSFP56 QSFP112 QSFP112 / OSFP
Maximum Port Configuration Up to 2 × 100 Gb/s Dual-port configurations vary by model Up to 2 × 200 Gb/s Model dependent Up to 2 × 400 Gb/s Model / configuration dependent
RDMA / RoCE ✓ Supported ✓ Supported ✓ Supported
GPUDirect RDMA ✓ Supported ✓ Supported ✓ Supported
Secure Boot Available on selected SKUs Available on selected SKUs Available on selected SKUs
Crypto Acceleration Available on selected SKUs Available on selected SKUs Available on selected SKUs
Typical Use Cases Data center networking, HPC, storage, virtualization AI/HPC clusters, cloud data centers, high-performance networking AI infrastructure, hyperscale data centers, HPC, next-generation networking
Generation Focus 200G-class networking 400G-class networking 400G Ethernet / 800G InfiniBand

Choosing the Right ConnectX NIC

For AI clusters, prioritize ConnectX-7 or ConnectX-8 for GPUDirect RDMA.
Choose 100GbE+ adapters for modern virtualization and storage workloads.
PCIe Gen5 and Gen6 network adapters provide significantly higher throughput for modern AI servers, GPU clusters, and high-bandwidth enterprise workloads.
InfiniBand is ideal for HPC and ultra-low latency compute environments.
Low-latency networking adapters help reduce communication bottlenecks between GPUs, storage systems, and compute nodes in AI and HPC environments

Frequently Asked Questions

What is ConnectX used for?

ConnectX NICs are used for AI, HPC, virtualization, cloud networking, and storage systems.

What is the difference between ConnectX-6, 7, and 8?

Each generation improves bandwidth, PCIe speed, and AI networking acceleration features.

What is VPI on NVIDIA ConnectX adapters?

VPI (Virtual Protocol Interconnect) allows supported NVIDIA ConnectX adapters to operate using both Ethernet and InfiniBand networking protocols. This provides flexbility for AI clusters, HPC environments, enterprise storage, and mixed networking infrastructures.

Do these support Ethernet and InfiniBand?

Yes, depending on the model, ConnectX adapters support both Ethernet and InfiniBand.

What speeds do NVIDIA ConnectX adapters support?

NVIDIA ConnectX adapters are available in multiple speed configurations, including 25GbE, 50GbE, 100GbE, 200GbE, and 400GbE depending on the model.

Are ConnectX adapters good for virtualization and storage networking?

ConnectX NICs are widely used in VMware environments, software-defined storage, NVMe-over-Fabrics, and enterprise virtualization platforms due to their advanced offloading and RDMA capabilities.

What is the difference between NVIIDA and Mellanox networking adapters?

Mellanox networking products are now part of NVIDIA networking following NVIDIA's acquisition of Mellanox Technologies. Many ConnectX adapters are still commonly referred to as Mellanox NICs, although they are now branded under NVIDIA Networking.