Datasheet

ConnectX-7
400G Adapters
Accelerated networking for modern
data center infrastructures.

The NVIDIA® ConnectX®-7 family of remote direct-memory access (RDMA)

Product Specifications
network adapters supports InfiniBand and Ethernet protocols and a range of
Supported InfiniBand,
speeds up to 400 gigabits per second (Gb/s). It enables a wide range of advanced,
network Ethernet
scalable, and secure networking solutions that address traditional enterprise needs protocols
up to the world’s most demanding AI, scientific computing, and hyperscale cloud
Total bandwidth 400Gb/s
data center workloads.
InfiniBand NDR 400Gb/s, HDR
speeds 200Gb/s, EDR 100Gb/s
Accelerated Networking and Security
Ethernet speeds 400GbE, 200GbE,
ConnectX-7 provides a broad set of software-defined, hardware-accelerated 100GbE, 50GbE,
networking, storage, and security capabilities, which enable organizations to 25GbE, 10Gb

modernize and secure their IT infrastructures. ConnectX-7 also powers agile and Number of 1, 2, or 4
high-performance solutions from edge to core data centers to clouds, all while network ports

enhancing network security and reducing total cost of ownership. Host interface PCIe Gen5,
up to 32 lanes

Accelerate Data-Driven Scientific Computing Form factors PCIe HHHL/FHHL and

OCP3.0 TSFF/SFF
ConnectX-7 provides ultra-low latency, extreme throughput, and innovative
NVIDIA In-Network Computing engines to deliver the acceleration, scalability, and Interface NRZ (10G, 25G),
technologies PAM4 (50G, 100G)
feature-rich technology needed for today’s modern scientific computing workloads.

Features*
InfiniBand Interface Ethernet Interface

> InfiniBand Trade Association Spec 1.5 > Up to four network ports supporting
compliant NRZ, PAM4 (50G and 100G),
in various configurations
> RDMA, send/receive semantics
> Up to 400Gb/s total bandwidth
> 16 million input/output (IO) channels
> RDMA over converged Ethernet
> 256 to 4 kilobyte maximum
(RoCE)
transmission unit (MTU), 2Gb
messages

NVIDIA ConnectX-7 400G Adapters | Datasheet | 1

Enhanced InfiniBand Networking Enhanced Ethernet Networking

> Hardware-based reliable transport > Zero-touch RoCE

> Extended Reliable Connected (XRC) > NVIDIA Accelerated Switch and
Packet Processing (ASAP2)™ for
> Dynamically Connected Transport software-defined networking (SDN)
(DCT) and virtual network functions (VNF)
> NVIDIA GPUDirect® RDMA > Open vSwitch (OVS) acceleration
> GPUDirect Storage > Overlay network acceleration:
Virtual Extensible LAN (VXLAN),
> Adaptive routing support Generic Network Virtualization
Encapsulation (GENEVE), Network
> Enhanced atomic operations
Virtualization using Generic
> Advanced memory mapping, allowing Routing Encapsulation (NVGRE)
user mode registration (UMR) > Connection tracking (L4 firewall)
and network address translation
> On-demand paging (ODP), including
(NAT)
registration-free RDMA memory
access > Flow mirroring, header rewrite,
hierarchical quality of service (QoS)
> Enhanced congestion control
> Single-root IO virtualization (SR-IOV)
> Burst buffer offload
> Stateless Transmission Control
> Single root IO virtualization (SR-IOV) Protocol (TCP) offloads

> Optimized for high-performance

Storage Accelerations
computing (HPC) software libraries,
including: > Block-level encryption: XEX-based
> NVIDIA HPC-X®, NVIDIA Unified tweaked codebook mode with
Communication X (UCX®), NVIDIA ciphertext stealing-Advanced
Unified Collective Communication Encryption Standard (XTS-AES)
(UCC), NVIDIA Collective 256/512-bit key
Communications Library (NCCL),
OpenMPI, MVAPICH, MPICH, > Non-Volatile Memory Express over
OpenSHMEM, partitioned global Fabrics (NVMe-oF) and NVMe/TCP
address space (PGAS) acceleration

> Collective operations offloads > T10 Data Integrity Field (T10-DIF)
signature handover
> Support for NVIDIA Scalable
Hierarchical Aggregation and > Server Routing Protocol (SRP),
Reduction Protocol (SHARP)™ Internet Small Computer Systems
Interface (iSCSI) Extensions for
> Rendezvous protocol offload
RDMA (iSER), Network File System
> In-network on-board memory (NFS) over RDMA, Server Message
Block (SMB) Direct
Remote Boot

> Remote boot over InfiniBand

> Remote boot over Internet Small

Computer Systems Interface (iSCSI)

> Unified Extensible Firmware Interface

(UEFI)

> Preboot Execution Environment (PXE)

NVIDIA ConnectX-7 400G Adapters | Datasheet | 2

Management and Control Advanced Timing and
Synchronization
> Network controller sideboard
interface (NC-SI), Management > Advanced Precision Time Protocol
Component Transport Protocol (PTP): IEEE 1588v2 (any profile),
(MCTP) over System Managment Bus G.8273.2 Class C, 12 nanosecond
(SMBus), accuracy, line-rate hardware
and MCTP over PCIe timestamp (UTC format)

> Platform Level Data Model (PLDM) for > SyncE: Meets G.8262.1 (eEEC)
Monitor and Control DSP0248
> Configurable packets per second
> PLDM for Firmware Update DSP0267 (PPS) in and out

> PLDM for Redfish Device Enablement > Time-triggered scheduling

DSP0218
> PTP-based packet pacing
> PLDM for Field-Replaceable Unit
(FRU) DSP0257 Compatibility
> Security Protocols and Data Models
PCI Express Interface
(SPDM) DSP0274
> PCIe Gen 5.0 compatible, 32 lanes
> Serial Peripheral Interface (SPI) to flash
> Support for PCIe bifurcation
> Joint Test Action Group (JTAG)
Institute of Electrical and Electronics > NVIDIA Multi-Host™ supports
Engineers (IEEE) 1149.1 and IEEE 1149.6 connection of up to four hosts

Cybersecurity > PCIe switch Downstream Port

Containment (DPC)
> Platform security:
> Support for Message Signaled
> Secure boot with hardware
Interrupts (MSI)/MSI-X mechanisms
root of trust
> Secure firmware update
> Flash encryption
> Device attestation

> Internet Protocol Security (IPsec)/

Transport Layer Security (TLS)/Media
Access Control Security (MACSec) Portfolio and Ordering
128/256-bit key data-in-motion
Information
encryption

> IPsec for RoCE and Ethernet The portfolio of ConnectX-7

network adapters and ordering
information is available in the
ConnectX-7 user manuals:
PCIe adapters manual and
OCP 3.0 adapters manual.
Ready to Get Started?
To learn more about InfiniBand adapters, visit:
nvidia.com/infiniband-adapters
To learn more about Ethernet network interface cards (NICs),
visit: nvidia.com/ethernet-adapters
© 2024 NVIDIA Corporation and affiliates. All rights reserved. NVIDIA, the NVIDIA logo, Accelerated Switch and
Packet Processing (ASAP2), ConnectX, GPUDirect, HPC-X, Multi-Host, Scalable Hierarchical Aggregation and
Reduction Protocol (SHARP), and UCX are trademarks and/or registered trademarks of NVIDIA Corporation. All
company and product names are trademarks or registered trademarks of the respective owners with which they are
associated. Features, pricing, availability, and specifications are all subject to change without notice. 3083375. JAN24