Skip to content

Cnuas Facility Twin, Datasheet

Experimental research preview

Facility Twin is in its infancy. Existing code provides an early exploratory prototype for scene generation, simplified load calculations and connections to emulated rack telemetry, not a ready or validated data-centre digital twin. Published is the status of this document, not a product-readiness or source-release certification. Do not use its outputs for engineering, procurement, safety or operational decisions.

Item Value
Part cnuas-facility
Type Experimental facility scene and simplified load prototype
Package cnuas-facility, version 0.1.0
Version 5c3d075-dirty
Repo PacketFive/cnuas

1. Overview

Cnuas is an open-source product/platform and an extensible research framework. The overall effort is experimental research, with some implemented prototypes and other capabilities in progress or planned. Its core and accelerator stack are intended for collaborative community development.

The existing Facility Twin prototype generates an illustrative AI and HPC data-centre campus as an OpenUSD scene, calculates simplified load roll-ups and writes readings from an emulated Open Rack V3 power shelf onto scene attributes. These are exploratory software interfaces, not evidence of accurate physical behaviour or a validated campus design.

We intend to make this early prototype available in the public release. This does not promise a finished Facility Twin extension in the initial release or mean that the full source has already been publicly released.

OpenUSD is the scene interchange format. NVIDIA Omniverse and Isaac Sim are intended exploration environments, not a certified compatibility matrix. Successful loading, materials, attributes and telemetry behaviour depend on the viewer and version; compatibility with every OpenUSD viewer is not claimed.

Implemented exploratory features

  • Declarative illustrative campus specification, from site dimensions to racks.
  • Strict user JSON configuration with template generation and spatial checks before OpenUSD authoring, not validation of an engineering design.
  • Three simplified building footprints, the I, H and E shapes.
  • Geometric data halls with rack rows, aisle spacing and support rooms.
  • Illustrative yards, roads, plant compounds, groves and lakes.
  • Assumed-load roll-up from rack to campus, with illustrative plant counts.
  • Sampling of the emulated Open Rack V3 shelf over Modbus, written onto the scene as typed attributes; physical-system fidelity remains unvalidated.

2. Module map

Module Responsibility
spec.py Declarative campus, building and hall specification, and the illustrative reference campus
config.py JSON template, strict loader, custom rack catalogue and spatial checks
catalog.py Assumed equipment dimensions and load ratings per rack and plant type
plan.py Resolves a specification into placed rooms, racks, plant and roads
layout.py Rectangle mathematics, building footprints, rack row layout
geom.py Geometry and material primitives
usd.py Writes the campus OpenUSD scene graph
schema.py The cnuas: attribute namespace carried on the scene
power.py Simplified load roll-up and illustrative plant counts
orv3.py Modbus client and register decoding for the power shelf
live.py Samples the shelf and animates the scene
units.py Named dimensional constants
cli.py Command line front end

3. Command reference

Command Purpose
build Write the campus OpenUSD stage
power Print the power roll-up
rooms Print the room programme
live Sample the Open Rack V3 shelf and animate the stage
template Write the complete reference campus as editable JSON
validate Check custom JSON structure and spatial constraints without writing OpenUSD; not physical scenario validation

4. Scene specification

These reference values are illustrative inputs or targets, not a measured facility, calibrated design or demonstrated operating scenario.

Parameter Reference value
Campus name Cnuas Ridge AI Campus
Site footprint 1000 m by 800 m
Buildings 3, one each of the I, H and E footprints
Data halls 16 across the three buildings
Assumed power usage effectiveness 1.15
Illustrative target campus load Close to 300 MW
Stage root /World
Scene branches /World/Looks, /World/Site, /World/Yards, /World/Buildings, /World/Prototypes
Default output format .usda, with .usdc and .usd also supported

5. Equipment catalogue

Entry Assumed load or sizing basis
Cnuas Open Rack V3, 24 OpenU 6 kW
GB200 NVL72 class rack 132 kW
Storage rack 12 kW
Network rack 15 kW
Medium voltage switchgear, transformer, generator, coolant distribution unit, fan wall, indirect evaporative cooler, dry cooler Simplified catalogue and power-model assumptions

6. Power model

The arithmetic below is implemented. It uses simplified illustrative assumptions, not a calibrated electrical or thermal simulation. Catalogue ratings, PUE, heat rejection and plant unit sizes must not be interpreted as validated equipment specifications or used for engineering or operational decisions.

Quantity Definition
Information technology load Sum of rack ratings in scope
Facility load Information technology load multiplied by the power usage effectiveness
Overhead Facility load less information technology load
Heat rejected Assumed equal to the information technology load
Density Load per unit floor area

Roll-up functions run at hall, building and campus scope. Illustrative plant counts are derived from the resulting assumed load; they do not establish capacity adequacy, redundancy, protection or cooling performance.

Plant Assumed unit size used for illustrative counts
Generator 3000 kW
Cooler 3500 kW
Coolant distribution unit 2000 kW

7. Live telemetry

Item Value
Source Emulated Open Rack V3 power shelf
Protocol Modbus remote terminal unit framing
Transport Serial line or a TCP endpoint, by default port 3485 on the loopback address
Decoded from the power supply Voltage, current, power, fan speed, temperature
Decoded from the battery backup State of charge and unit state
Write back Typed attributes in the cnuas: namespace on the corresponding scene primitives

The shelf itself is emulated by the BMC stack, which runs the power shelf firmware and presents the RS-485 segment.

These connections exercise software data paths from emulated telemetry. They do not establish fidelity to physical shelves, end-to-end fault coverage or validated coupling between rack behaviour and campus electrical or thermal behaviour.

8. Software checks and outstanding scenario validation

The following inventory covers prototype software tests under facility/. Passing these checks does not validate facility scenarios, physics, design suitability or product readiness.

Area Test Cases
Power roll-up and plant sizing tests/test_power.py 8
Footprints, rectangles and rack rows tests/test_layout.py 19
Scene graph generation, including custom-configuration read-back tests/test_usd.py 14
Shelf sampling and scene animation tests/test_live.py 19
Custom JSON round trip, validation, automatic plant sizing and CLI tests/test_config.py 6

The collected inventory totals 66 software cases. See the Validation Matrix for software-check context, not evidence of a validated facility twin.

Substantial scenario validation remains needed. The following work is not claimed complete.

Area requiring validation Work still needed
Geometry and catalogue assumptions Check dimensions, placement constraints, aisle and plant clearances, equipment representations and applicable reference data
Electrical, load and thermal behaviour Compare assumed rack loads, PUE, heat rejection and plant counts with reference calculations and measurements across operating conditions
Telemetry integration and faults Assess mapping, units, timing, stale or missing readings, malformed data, disconnects and fault propagation across the integrated scenario
References and calibration Establish traceable datasets, calibration procedures, uncertainty bounds and acceptance criteria
Held-out scenarios Evaluate independent scenarios not used to select assumptions or calibrate models, including abnormal and boundary conditions

9. Operating notes

  • The prototype represents geometry, simplified load and emulated telemetry. It does not perform computational fluid dynamics, electrical fault studies or structural analysis, and the plant counts are illustrative arithmetic rather than engineering designs.
  • Building footprints are limited to the I, H and E shapes in this revision.
  • Custom configurations can add rack types, but plant item types remain the built-in electrical and mechanical catalogue in this revision.

10. Integration information

Item Value
Repo PacketFive/cnuas
Source facility/src/cnuas_facility/
Entry point cnuas-facility
Python 3.10 or later
Dependencies OpenUSD Python bindings
Viewers OpenUSD-capable tools; Omniverse and Isaac Sim are intended exploration environments, with viewer/version compatibility requiring validation
Release intent Make the existing early prototype available publicly, not deliver a finished Facility Twin extension in the initial release

11. Revision history

Revision Notes
A First publication. Reference campus, catalogue ratings and plant unit sizes read from the source.
B Added the beginner workflow and strict custom-campus JSON template, validation, custom rack types, configurable roads and automatic per-building plant counts.
C Marked experimental research preview, public-release intent and limits of software checks, load assumptions, viewer compatibility and scenario validation.