GPU Expansion Rack Configuration

Private Cloud Appliance with GPU expansion provides a scalable platform to build AI and graphics intensive applications in the private cloud environment.

It's built to power the next generation of data center workloads, including:

  • Generative AI inference: real time inferencing for multimodel generative AI pipelines (text, image, audio, video)

  • LLM training and fine-tuning: accelerated performance for fine-tuning medium LLMs and training small LLMs with NVIDIA's transformer engine and FP8 support

  • Graphics-intensive and VDI applications: 3D graphics and rendering workflows with NVIDIA’s RTX and ray tracing capabilities

  • Digital twins using NVIDIA Omniverse: develop and operate complex 3D industrial digitization workflows

  • Media streaming: increased encode/decode density and AV1 support for 4K video streaming

  • HPC: scientific data analysis and simulation workloads with FP32 support

GPU expansions require appliance software version 3.0.2-b1325160 (March 2025) or newer. An X10-2c GPU expansion rack contains 1 to 6 X10-2c GPU L40S Compute Server nodes. To integrate with the base rack physical network infrastructure, two Cisco Nexus 9336C-FX2 leaf switches and a Cisco Nexus 9348GC-FXP management switch are installed. This rack does not include storage hardware.

X10-2c GPU Rack Configuration

The minimum configuration adds 1 factory-installed GPU expansion node. More nodes can be installed at the factory or after deployment. Cabling is preinstalled for a full rack configuration, regardless of the number of factory-installed nodes. A single expansion rack contains up to 6 GPU nodes. Two expansion racks can be connected to the base rack, for a maximum of 12 GPU nodes.


Figure showing the components installed in a GPU expansion rack.

Callout

Quantity

Description

A

6

GPU node

minimum configuration: 1, rack maximum: 6

B

1

brush filler - allows cable routing from the back to connectors in the front

C

1

management switch

D

2

universal power distribution unit (UPDU)

E

2

leaf switch

(none)

18

filler panel

installed in the top 12 rack units and in empty spaces between components