Introduction

Data Rate

InfiniBand: SDR/DDR/QDR/FDR
Ethernet: 1/10/25/40/56 Gb/s

Network Connector Type

Single-port QSFP28

Dual-port QSFP28

PCI Express Connectors

PCIe 3.0 x8 @ 8GT/s

PCIe 3.0 @ 8GT/s x16

PCIe 3.0 x8 @ 8GT/s

PCIe 3.0 @ 8GT/s x16

Dimensions

2.71 in. x 5.6 in. (68.90mm x 142.24 mm) – low profile

RoHS

RoHS Compliant

Adapter IC Part Number

MT27704A0-FDCF-FV

Device ID (decimal)

4115 for Physical Function (PF)
4116 for Virtual Function (VF)

Data Rate

InfiniBand: SDR/DDR/QDR/FDR/EDR
Ethernet: 10/25/40/50/100 Gb/s

Network Connector Type

Single-port QSFP28

Dual-port QSFP28

PCI Express Connectors

PCIe Gen 3.0 x16; SerDes @ 8.0GT/s

Dimensions

2.71 in. x 5.6 in. (68.90mm x 142.24 mm) – low profile

RoHS

RoHS Compliant

Adapter IC Part Number

MT27704A0-FDCF-EV

Device ID (decimal)

4115 for Physical Function (PF)
4116 for Virtual Function (VF)

PCI Express (PCIe)

Uses PCIe Gen 3.0 (8GT/s) through an x8 or x16 edge connector. Gen 1.1 and 2.0 compatible.

EDR InfiniBand

A standard InfiniBand data rate, where each lane of a 4X port runs a bit rate of 25.78125Gb/s with a 64b/66b encoding, resulting in an effective bandwidth of 100Gb/s.

100Gb/s Adapter

ConnectX-4 offers the highest throughput InfiniBand/Ethernet adapter, supporting EDR 100Gb/s InfiniBand and 100Gb/s Ethernet and enabling any standard networking, clustering, or storage to operate seamlessly over any converged network leveraging a consolidated software stack.

InfiniBand Architecture Specification v1.3 compliant

ConnectX-4 delivers low latency, high bandwidth, and computing efficiency for performance-driven server and storage clustering applications. ConnectX-4 is InfiniBand Architecture Specification v1.3 compliant.

Up to 100 Gigabit Ethernet

NVIDIA adapters comply with the following IEEE 802.3 standards:

• 100GbE/ 50GbE / 40GbE / 25GbE / 10GbE / 1GbE

• IEEE 802.3bj, 802.3bm 100 Gigabit Ethernet

• IEEE 802.3by, Ethernet Consortium25, 50 Gigabit Ethernet, supporting all FEC modes

• IEEE 802.3ba 40 Gigabit Ethernet

• IEEE 802.3ae 10 Gigabit Ethernet

• IEEE 802.3ap based auto-negotiation and KR startup

• Proprietary Ethernet protocols (20/40GBASE-R2, 50GBASE-R4)

• IEEE 802.3ad, 802.1AX Link Aggregation

• IEEE 802.1Q, 802.1P VLAN tags and priority

• IEEE 802.1Qau (QCN)

• Congestion Notification

• IEEE 802.1Qaz (ETS)

• IEEE 802.1Qbb (PFC)

• IEEE 802.1Qbg

• IEEE 1588v2

• Jumbo frame support (9.6KB)

Memory

• SPI - includes one16MB SPI Flash device(M25PX16-VMN6P device by ST Microelectronics)

• FRU EEPROM - Stores the parameters and personality of the card. The EEPROM capacity is 128Kbit. FRU I2C address is (0x50) and is accessible through the PCIe SMBus. (Note: Address 0x58 is reserved.)

Overlay Networks

In order to better scale their networks, data center operators often create overlay networks that carry traffic from individual virtual machines over logical tunnels in encapsulated formats such as NVGRE and VXLAN. While this solves network scalability issues, it hides the TCP packet from the hardware offloading engines, placing higher loads on the host CPU. ConnectX-4 effectively addresses this by providing advanced NVGRE and VXLAN hardware offloading engines that encapsulate and de-capsulate the overlay protocol.

RDMA and RDMA over Converged Ethernet (RoCE)

ConnectX-4, utilizing IBTA RDMA (Remote Data Memory Access) and RoCE (RDMA over Converged Ethernet) technology, delivers low-latency and high performance over Band and Ethernet networks. Leveraging data center bridging (DCB) capabilities as well as ConnectX-4 advanced congestion control hardware mechanisms, RoCE provides efficient low-latency RDMA services over Layer 2 and Layer 3 networks.

NVIDIA® PeerDirect™

PeerDirect™ communication provides high-efficiency RDMA access by eliminating unnecessary internal data copies between components on the PCIe bus (for example, from GPU to CPU), and therefore significantly reduces application run time. ConnectX-4 advanced acceleration technology enables higher cluster efficiency and scalability to tens of thousands of nodes.

CPU Offload

Adapter functionality enabling reduced CPU overhead allowing more available CPU for computation tasks.

Quality of Service (QoS)

Support for port-based Quality of Service enabling various application requirements for latency and SLA.

Hardware-based I/O Virtualization

ConnectX-4 provides dedicated adapter resources and guaranteed isolation and protection for virtual machines within the server.

Storage Acceleration

A consolidated compute and storage network achieves significant cost-performance advantages over multi-fabric networks. Standard block and file access protocols can leverage InfiniBand RDMA for high-performance storage access.

• NVMe over Fabric offloads for the target machine

SR-IOV

ConnectX-4 SR-IOV technology provides dedicated adapter resources and guaranteed isolation and protection for virtual machines (VM) within the server.