This site requires Javascript in order to view all its content. Please enable Javascript in order to access all the functionality of this web site. Here are the instructions how to enable JavaScript in your web browser.

NVIDIA Mellanox ConnectX-6 VPI

The World’s First HDR 200Gb/s InfiniBand Host Channel Adapter

Intelligent NVIDIA^® Mellanox^® ConnectX^®-6 adapter cards deliver high performance and NVIDIA In-Network Computing acceleration engines for maximizing efficiency in high-performance computing (HPC), artificial intelligence (AI), cloud, hyperscale, and storage platforms.

Single/Dual-Port Adapter Supporting 200Gb/s

ConnectX-6 Virtual Protocol Interconnect^® (VPI) adapter cards offer up to two ports of 200Gb/s throughput for InfiniBand and Ethernet connectivity, provide ultra-low latency, deliver 215 million messages per second, and feature innovative smart offloads and in-network computing accelerations that drive performance and efficiency.

ConnectX-6 is a groundbreaking addition to the ConnectX series of industry-leading adapter cards, providing innovative features such as in-network memory capabilities, message passing interface (MPI) tag matching hardware acceleration, out-of-order RDMA write and read operations, and congestion control over HDR, HDR100, EDR, and FDR InfiniBand speeds.

LEARN MORE ›

NVIDIA Mellanox Dual-Port 100GbE/Single-Port 200GbE SmartNIC

Highlights

PORT SPEEDS

2x200Gb/s

TOTAL BANDWIDTH

200Gb/s

MESSAGE RATE

215 million msgs/sec

PCIe LANES

32x Gen3/Gen4

Use Case Applications

High-Performance Computing (HPC)

ConnectX-6 delivers the highest throughput and message rate in the industry and is the perfect product to lead HPC data centers toward exascale levels of performance and scalability.

ConnectX-6 offers enhancements to HPC infrastructures by providing MPI acceleration and offloading, as well as support for network atomic and PCIe atomic operations.

Machine Learning and AI

Machine learning relies on high throughput and low latency to train deep neural networks and to improve recognition and classification accuracy. As the first adapter card to deliver 200Gb/s throughput supporting NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) , ConnectX-6 and Quantum switches provide machine learning applications with the performance and scalability they need.

Storage

NVMe storage devices are gaining momentum by offering exceptionally fast remote direct-memory access (RDMA) to storage media. With its NVMe over Fabrics (NVMe-oF) target and initiator offloads, ConnectX-6 brings further optimizations to NVMe-oF, enhancing CPU utilization and scalability.

Key Features

HDR / HDR100 / EDR / FDR / QDR / SDR InfiniBand connectivity
Up to 200Gb/s bandwidth
Up to 215 M messages per second
Low latency
RDMA, send/receive semantics
Hardware-based congestion control
Atomic operations

Collective operations offloads
Support for NVIDIA Mellanox Multi-Host^® and NVIDIA Mellanox Socket Direct^® configurations
Embedded PCIe switch
Available in a variety of form factors: PCIe standup, NVIDIA Socket Direct, OCP3.0, NVIDIA Multi-Host, and integrated circuit (IC) standalone

Benefits

Delivers the highest throughput and message rate in the industry
Offloads computation to save CPU cycles and increase network efficiency
Highest performance and most intelligent fabric for compute and storage infrastructures
Support for x86, Power, Arm, and GPU-based compute and storage platforms