NVIDIA Networking InfiniBand Solutions

NVIDIA InfiniBand Products

You are looking for Gigabit Ethernet solutions? Then check out our selection of Nvidia GbE Switches and Adapters.

InfiniBand Switch Systems

The NVIDIA's family of InfiniBand switches deliver the highest performance and port density with complete fabric management solutions to enable compute clusters and converged data centers to operate at any scale while reducing operational costs and infrastructure complexity. Nvidia switches includes a broad portfolio of Edge and Director switches supporting 40, 56, 100, 200 and 400Gb/s port speeds and ranging from 8 to 800 ports.

Edge Switches

The Edge family of switch systems provide the highest-performing fabric solutions in a 1U form factor by delivering up to 51Tb/s of non-blocking bandwidth with the lowest port-to-port latency. These edge switches are an ideal choice for top-of-rack leaf connectivity or for building small to medium sized clusters. The edge switches, offered as externally managed or as managed switches, are designed to build the most efficient switch fabrics through the use of advanced InfiniBand switching technologies such as Adaptive Routing, Congestion Control and Quality of Service.

NVIDIA SWITCH-IB 2 SERIES	MSB7880	MSB7890
Type	Router	Switch
SHARP	SHARP v1	SHARP v1
Ports	36	36
Height	1U	1U
Switching Capacity	7.2Tb/s	7.2Tb/s
Link Speed	100Gb/s	100Gb/s
Interface Type	QSFP28	QSFP28
Management	Yes (internally)	No (only externally)
Management Ports	1x RJ45, 1x RS232, 1x USB	1x RJ45
Power	1+1 redundant and hot-swappable, 80 Gold+ and ENERGY STAR certified
System Memory	4GB RAM DDR3
Storage	16GB SSD
Cooling	Front-to-rear or rear-to-front (hot-swappable fan unit)

NVIDIA QUANTUM SERIES	MQM8700	MQM8790	MQM9700	MQM9790
Series	Quantum	Quantum	Quantum-2	Quantum-2
SHARP	SHARP v2	SHARP v2	SHARP v3	SHARP v3
Ports	40	40	64	64
Height	1U	1U	1U	1U
Switching Capacity	16Tb/s	16Tb/s	51.2Tb/s	51.2Tb/s
Link Speed	200Gb/s	200Gb/s	400Gb/s	400Gb/s
Interface Type	QSFP56	QSFP56	QSFP	QSFP
Management	Yes (internally)	No (only externally)	Yes (internally)	No (only externally)
Management Ports	1x RJ45, 1x RS232, 1x micro USB		1x USB3.0, 1x USB for I2C, 1x RJ45, 1x RJ45(UART)
Power	1+1 redundant and hot-swappable, 80 Gold+ and ENERGY STAR certified
System Memory	Single 8GB		Single 8GB DDR4 SO-DIMM
Storage	---		M.2 SSD SATA 16GB 2242 FF
Cooling	Front-to-rear or rear-to-front (hot-swappable fan unit)

Modular Switches

Nvidia’s director modular switches provide the highest density switching solution, scaling from 8.64Tb/s up to 320Tb/s of bandwidth in a single enclosure, with low-latency and the highest per port speeds of up to 200Gb/s. Its smart design provides unprecedented levels of performance and makes it easy to build clusters that can scale out to thousands-of-nodes. The InfiniBand modular switches deliver director-class availability required for mission-critical application environments. The leaf, spine blades and management modules, as well as the power supplies and fan units, are all hot-swappable to help eliminate down time.

managed 324 - 800 ports	CS7520	CS7510	CS7500	CS8500
Ports	216	324	648	324
Height	12U	16U	28U	29U
Switching Capacity	43.2Tb/s	64.8Tb/s	130Tb/s	320Tb/s
Link Speed	100Gb/s	100Gb/s	100Gb/s	200Gb/s
Interface Type	QSFP28	QSFP28	QSFP28	QSFP28
Management	2048 nodes	2048 nodes	2048 nodes	2048 nodes
Management HA	Yes	Yes	Yes	Yes
Console Cables	Yes	Yes	Yes	Yes
Spine Modules	6	9	18	20
Leaf Modules (max.)	6	9	18	20
PSU Redundancy	Yes (N+N)	Yes (N+N)	Yes (N+N)	Yes (N+N)
Fan Redundancy	Yes	Yes	Yes	Water cooled

Benefits Nvidia Switch Systems

• Built with Nvidia's 4th and 5th generation InfiniScale® and SwitchX™ switch silicon
• Industry-leading energy efficiency, density, and cost saving switches company
• Ultra low latency
• Real-Time Scalable Network Telemetry
• Scalability and subnet isolation using InfiniBand routing and InfiniBand to Ethernet gateway capabilities
• Granular QoS for Cluster, LAN and SAN traffic
• Quick and easy setup and management
• Maximizes performance by removing fabric congestions
• Fabric Management for cluster and converged I/O applications

ConnectX-7 VPI Cards

As a key component of the NVIDIA® Quantum-2 InfiniBand platform, the NVIDIA ConnectX®-7 smart host channel adapter (HCA) provides the highest networking performance available to take on the world’s most challenging problems. The ConnectX-7 InfiniBand adapter provides ultra-low latency, 400Gb/s throughput, and innovative NVIDIA In-Network Computing engines to deliver the acceleration, scalability, and feature-rich technology needed for high performance computing (HPC), artificial intelligence (AI), and hyperscale cloud data centers.

ConnectX-7 (400 Gb/s)	MCX75310AAS-NEAT	MCX75310AAS-HEAT	MCX755106AS-HEAT	MCX75343AAS-NEAC	MCX753436AS-HEAB
ASIC & PCI Dev ID	ConnectX®-7
Form Factor	PCIe Standup Form Factor (HHHL)			OCP 3.0 (TSFF)	OCP 3.0 (SFF)
Ports	1	2	1	1	2
Speed	400Gb/s	200Gb/s	200Gb/s	400Gb/s	200Gb/s
Connectors	1x QSFP	1x QSFP	2x QSFP	1x QSFP	2x QSFP *5
PCI	PCIe 4.0/5.0	PCIe 4.0/5.0	PCIe 4.0/5.0	PCIe 4.0/5.0	PCIe 4.0/5.0
Lanes	x16 *1*2	x16 *3*4	x16 With option for extension	x16	x16
RDMA message rate	330-370 million messages per second
Dimensions	without brackets: 167.65mm x 68.90mm. All adapters are shipped with the tall bracket mounted and a short bracket as an accessory

^*1PCEe 4.0/5.0 x16 with option for extension is available with model: MCX75510AAS-NEAT
^*2PCEe 4.0/5.0 2x8 in a row is available with model: MCX75210AAS-NEAT
^*3PCEe 4.0/5.0 x16 with option for extension is available with model: MCX75510AAS-HEAT
^*4PCEe 4.0/5.0 2x8 in a row is available with model: MCX75210AAS-HEAT
^*5This card supports one port of InfiniBand, and a second port as either InfiniBand or Ethernet

ConnectX-6 VPI Cards

ConnectX-6 with Virtual Protocol Interconnect® (VPI) supports two ports of 200Gb/s InfiniBand and Ethernet connectivity, sub-600 nanosecond latency, and 200 million messages per second, providing the highest performance and most flexible solution for the most demanding applications and markets. Delivering one of the highest throughput and message rate in the industry with 200Gb/s HDR InfiniBand, 100Gb/s HDR100 InfiniBand and 200Gb/s Ethernet speeds it is the perfect product to lead HPC data centers toward Exascale levels of performance and scalability. Supported speeds are HDR, HDR100, EDR, FDR, QDR, DDR and SDR InfiniBand as well as 200, 100, 50, 40, 25, and 10Gb/s Ethernet. All card speeds are backwards compatible.

ConnectX-6 (200 Gb/s)	MCX654105A-HCAT	MCX654106A-HCAT	MCX653105A-HDAT	MCX653106A-HDAT	MCX683105AN-HDAT
ASIC & PCI Dev ID	ConnectX®-6				ConnectX®-6 DE
Speed	HDR IB (200Gb/s) & 200GbE, HDR100, EDR, FDR, QDR, DDR, SDR				HDR IB (200Gb/s)
Ports	1	2	1	2	1
Connectors	QSFP56	QSFP56	QSFP56	QSFP56	QSFP
PCI	2x PCIe 3.0 (Socket Direkt)	2x PCIe 3.0 (Socket Direkt)	PCIe 3.0/4.0	PCIe 3.0/4.0	PCIe 3.0/4.0
Lanes	x16	x16	x16	x16	x16
Bracket	tall bracket*
Dimensions	PCIe Standup form factor. w/o Bracket: 167.65mm x 8.90mm

ConnectX-6 (100 Gb/s)	MCX654106A-ECAT	MCX653105A-ECAT	MCX653106A-ECAT
ASIC & PCI Dev ID	ConnectX®-6
Speed	HDR100 (100Gb/s), EDR IB & (100GbE), FDR, QDR, DDR, SDR
Ports	2	1	2
Connectors	QSFP56	QSFP56	QSFP56
PCI	2x PCIe 3.0	PCIe 3.0/4.0	PCIe 3.0/4.0
Lanes	x16	x16 (x8*1) (2x8*2)	x16 (2x8*3)
Form Factor	PCIe Standup. w/o Bracket: 167.65mm x 8.90mm

^*1PCEe 4.0 x8 available with model: MCX651105A-EDAT
^*2PCEe 3.0/4.0 x16 Socket Direct 2x8 in a row, available with model: MCX653105A-EFAT
^*3PCEe 3.0/4.0 x16 Socket Direct 2x8 in a row, available with model: MCX653106A-EFAT

ConnectX-6	MCX653435A-HDAI	MCX653436A-HDAI	MCX653435A-EDAI	MCX653435A-HDAE
ASIC & PCI Dev ID	ConnectX®-6
Speed	HDR IB (200Gb/s) & 200GbE		HDR100 (100Gb/s) & 100GbE	HDR IB (200Gb/s) & 200GbE
Ports	1	2	1	1
Connectors	QSFP56	QSFP56	QSFP56	QSFP56
PCI	PCIe 3.0/4.0	PCIe 3.0/4.0	PCIe 3.0/4.0	PCIe 3.0/4.0
Lanes	x16	x16	x16	x16
Form Factor	OCP 3.0 Small Form Factor

ConnectX-5 VPI Cards

ConnectX-5 with Virtual Protocol Interconnect® supports two ports of 100Gb/s InfiniBand and Ethernet connectivity, sub-600ns latency, and very high message rate, plus PCIe switch and NVMe over Fabric offloads, providing the highest performance and most flexible solution for the most demanding applications and markets: Machine Learning, Data Analytics, and more.

ConnectX-5	MCX555A-ECAT	MCX556A-ECAT	MCX556A-EDAT	MCX545A-ECAN	MCX545M-ECAN
ASIC & PCI Dev ID	ConnectX®-5			ConnectX®-5 for OCP
Speed	EDR IB (100Gb/s) & 100GbE
Ports	1	2	2	1	1
Connectors	QSFP28	QSFP28	QSFP28	QSFP28	QSFP28
PCI	PCIe3.0	PCIe3.0	PCIe4.0	PCIe3.0	2x PCIe3.0
Lanes	x16	x16	x16	x16	x16
Bracket	tall bracket*			no bracket
Dimensions	14.2cm x 6.9cm (Low Profile)			OCP 2.0 type 2**

NVIDIA DPU (Data Processing Unit)

The NVIDIA^® BlueField^® Data Processing Unit (DPU) is a system on a chip, a hardware accelerator specialized for complex tasks like fast data processing and data-centric computing. This is primarily intended to relieve the CPU of network and communication tasks and thus save CPU resources by outsourcing and taking over application-supporting tasks such as data transfer, data reduction, data security and analyses. The use of a DPU is particularly suitable for all workloads with supercomputing tasks such as AI, cloud and big data. A dedicated operating system on the chip can be combined with the primary operating system and offers functions such as encryption, erasure coding and compression or decompression.

BlueField®-2 IB DPUs	MBF2M345A	MBF2H516A	MBF2H516C	MBF2M516A	MBF2M516C
Series / Core Speed	E-Series / 2.0GHz	P-Series / 2.75GHz		E-Series / 2.0GHz
Form Factor	Half-Height Half-Length (HHHL)	Full-Height Half-Length (FHHL)
Ports	1x QSFP56	2x QSFP56	2x QSFP56	2x QSFP56	2x QSFP56
Speed	200GbE / HDR	100GbE / EDR / HDR100
PCI	PCIe 4.0 x16
On-board DDR	16GB
On-board eMMC	64GB	64GB	128GB	64GB	128GB
Secure Boot	✓	-	✓	✓ (*1)	✓
Crypto	✓ (*1)	✓ (*1)	✓ (*1)	✓ (*1)	✓ (*1)
1GbE OOB	✓	✓	✓	✓	✓
Integrated BMC	-	-	✓	✓	✓
External Power	-	✓	✓	-	-
PPS IN/OUT	-	-	-	✓	✓

^*1depending on the specific postfix model name (see full list)