Name: NVIDIA Quantum-2 CS9500 Modular InfiniBand Switch
Brand: NVIDIA
SKU: MCS9500

NVIDIA

NVIDIA Quantum-2 CS9500 Modular InfiniBand Switch

A 29U modular NDR InfiniBand chassis family. 512, 1,024 or 2,048 x 400G ports, up to 1,600 Tb/s, liquid cooled; scales with leaf and spine blades.

Up to 2,048 x 400G NDR ports in one chassis (or 4,096 x 200G)
1,600 Tb/s aggregate bidirectional capacity
29U modular chassis, liquid cooled
Scales incrementally with leaf and spine blades

Models in this family

CS9500 MCS9500 CS9510 MCS9510 CS9520 MCS9520

Get a Quote +90 212 483 70 33

Authorised dealer supply · manufacturer warranty · price and lead time per configuration

The CS9500 is the modular-chassis counterpart of the Quantum-2 NDR generation. The fixed-configuration QM9700 family is ideal for the leaf layer, but in clusters of a few thousand nodes the spine layer means dozens of separate boxes and thousands of optical links. The CS9500 moves that layer inside a single 29U chassis.

There are three capacity tiers: MCS9520 at 512 ports, MCS9510 at 1,024, and MCS9500 at 2,048 ports of 400G NDR. Split down to 200G, those numbers double. The largest configuration reaches 1,600 Tb/s of aggregate bidirectional capacity.

The system is populated with leaf (MQM9510-N) and spine (MQM9520-N) blades, each carrying 128 ports over 64 OSFP cages. Because leaf-to-spine links are made across the chassis backplane, only the cables going to compute nodes leave the enclosure — most of the optical modules an equivalent fixed-switch topology would require simply disappear.

The chassis runs on closed-loop liquid cooling and is planned together with either a liquid-to-air (AHX) or liquid-to-liquid (CDU) unit, depending on your data centre. A CS9500-class system is therefore not a standalone hardware order but a project to be handled alongside mechanical plant, rack layout and power distribution.

Specifications

Specification	CS9500 MCS9500	CS9510 MCS9510	CS9520 MCS9520
Connectivity and capacity
NDR 400G ports	2,048	1,024	512
HDR 200G ports (split)	4,096	2,048	1,024
Aggregate capacity (bidirectional)	1,600 Tb/s	800 Tb/s	400 Tb/s
Connector	OSFP	OSFP	OSFP
Modular structure
Chassis	29U modular	29U modular	29U modular
Leaf blade	MQM9510-N — 128 ports / 64 OSFP, 2U	MQM9510-N — 128 ports / 64 OSFP, 2U	MQM9510-N — 128 ports / 64 OSFP, 2U
Spine blade	MQM9520-N — 128 ports / 64 OSFP, 2U	MQM9520-N — 128 ports / 64 OSFP, 2U	MQM9520-N — 128 ports / 64 OSFP, 2U
Hot-swap blades	Yes	Yes	Yes
Cooling and power
Cooling	Closed-loop liquid	Closed-loop liquid	Closed-loop liquid
Cooling option	Liquid-to-air (AHX) or liquid-to-liquid (CDU)	Liquid-to-air (AHX) or liquid-to-liquid (CDU)	Liquid-to-air (AHX) or liquid-to-liquid (CDU)
Power supplies	Hot-swap, redundant	Hot-swap, redundant	Hot-swap, redundant
Fabric capabilities
SHARPv3 (in-network collectives)	Yes	Yes	Yes
Adaptive routing	Yes	Yes	Yes
Congestion control	Yes	Yes	Yes
Onboard subnet manager	Yes	Yes	Yes

Specifications follow the manufacturer datasheet. The exact configuration is confirmed in your quote.

Connectivity Solutions

A spine layer in a single chassis

In a three-level fat-tree built from fixed-configuration switches, the spine layer means dozens of boxes and thousands of optical links. The CS9500 moves that layer inside one chassis: leaf-to-spine links are made across the chassis backplane, and only the cables going to compute nodes leave the enclosure.

Goes with InfiniBand transceivers

Incremental growth

The chassis is chosen for the final capacity you are targeting; blades are added as demand grows. If you plan to start at 512 ports and reach 2,048 over time, pick the chassis for that growth curve — swapping the chassis later is not a practical option.

Goes with InfiniBand switch models

Liquid cooling infrastructure

The system runs on closed-loop liquid cooling and is ordered together with either a liquid-to-air (AHX) or a liquid-to-liquid (CDU) unit, depending on your data centre. That choice has to be made alongside rack layout and mechanical plant.

Goes with HPC Infrastructure Deployment

Features

Exascale port density

2,048 ports of 400G in one 29U chassis is 2.5x the density of the previous modular generation. The rack space and the number of interconnect cables needed for the same compute capacity fall markedly.

Serviceable at blade level

Leaf blades, spine blades, management modules, power supplies and cooling units are all hot-swappable. A single component failure does not stop the fabric; maintenance can be done while the cluster runs.

Collective acceleration with SHARP

Collective MPI operations are reduced inside the chassis. In a modular design the benefit is even more pronounced, because most of the collective traffic already stays within one chassis.

Lower interconnect cost

Because leaf-spine links are made across the backplane, the thousands of optical modules and cables an equivalent fixed-switch topology would need disappear. That shrinks both the capital cost and the failure surface.