Platforms_
Getting Ready for Rubin / VR200: Deployment-Readiness Planning Now
Details the concrete facility, power, and cooling preparation steps operators must complete before Rubin-class racks arrive, including exact sequencing, standards references, and decision criteria for upgrades.
Key facts
- Next-generation GPU racks require liquid cooling loops separate from the facility chilled-water plant, with CDUs sized to the OEM rack thermal specification.
- Power feeds to each rack must be evaluated against the full rack draw stated in the OEM power summary before any breaker or busway work begins.
- TIA-942-B requires minimum clearance on both sides of a rack for liquid cooling service access; this dimension is non-negotiable for NVL-class hardware.
- ASHRAE TC 9.9 recommends supply air or liquid temperatures be verified against the OEM inlet specification rather than legacy air-cooled assumptions.
- MPO trunk cables for the scale-out fabric must be routed and tested independently of any internal copper NVLink backplane.
- Factory-terminated MPO connectors require cleaning and inspection with a calibrated scope before every mating; field polishing is not performed.
- Redundant power paths must be validated with a calibrated load bank at the planned rack draw before the first Rubin rack is placed.
Power Path Capacity Verification
Begin by obtaining the OEM rack power summary for the target Rubin configuration and compare it against existing busway or PDU ratings. Measure available amperage at the floor PDU and confirm that both A and B feeds can sustain the full load simultaneously. If either path falls short, schedule busway or transformer upgrades before rack delivery; attempting to add capacity after racks are on site creates schedule conflicts with live workloads.
Document the breaker coordination study and ensure the settings protect the rack PDUs without nuisance trips under the expected inrush from GPU power supplies. Perform this verification well in advance so utility coordination can occur if feeder upgrades are required.
Liquid Cooling Loop Sizing and CDU Placement
Map the facility chilled-water supply temperature and flow rate against the Rubin rack heat rejection requirement. Size each coolant distribution unit to handle one or two racks with headroom per OEM guidance; locate CDUs so that hose lengths stay within the OEM maximum to avoid excessive pressure drop. Route supply and return lines overhead or under raised floor with isolation valves at each rack position.
Confirm that the secondary loop uses the fluid specified by the rack OEM and that quick-disconnect fittings match the rack manifold. Test the entire loop for leaks at the pressure specified in the OEM installation guide before any GPUs are installed.
Rack Row Layout and Service Clearance
Re-evaluate row spacing using the TIA-942-B minimum side clearance requirement for liquid-cooled racks. This clearance must remain free of cable trays or containment panels so technicians can service manifolds and CDU connections without removing adjacent racks. Adjust containment or hot-aisle widths accordingly.
Place rack positions on the floor plan so that power whips and liquid hoses reach the correct rack PDU and manifold ports without sharp bends or tension. Verify that the resulting layout still satisfies any local fire-code aisle requirements.
Scale-Out Fabric Cabling Pathways
Design fiber pathways for the InfiniBand or Ethernet fabric separately from the internal copper NVLink spine. MPO trunks run between leaf and spine switches in adjacent racks or rows; maintain minimum bend radius throughout the route and reserve slack for future re-termination. Label every MPO connector with rack and port identifiers before installation.
Pull and test all trunks with an OTDR and a calibrated MPO continuity tester while the pathways are still accessible. This sequence prevents later discovery of damaged fibers after racks are powered.
Common Failure Modes During Pre-Deployment
The most frequent error is assuming existing chilled-water temperature will meet the new rack inlet specification; operators discover the mismatch only after CDUs are installed and racks arrive. Catch this by requesting the exact inlet temperature range from the OEM and measuring actual supply temperature at the planned CDU location.
Another common issue is under-sizing the secondary loop flow rate, which causes GPU throttling under sustained load. Validate flow with a temporary pump and flow meter before permanent piping is completed. Finally, power-path coordination studies are often skipped; the result is nuisance trips when all GPUs ramp simultaneously. Run a full load-bank test on both feeds before rack delivery.
Sequencing Validation and Commissioning
Complete power, cooling, and cabling work in that order, then perform integrated systems testing with load banks that simulate the Rubin rack profile. Confirm that leak detection, pressure alarms, and power redundancy all function before any production hardware is unboxed. Leviathan Systems schedules this validation window with the operator so that facility staff can witness every test point.
Only after all checks pass should racks be positioned and connected. This sequence eliminates the need to remove installed racks for corrective work.
Standards referenced: TIA-942-B · ASHRAE TC 9.9
Frequently asked_
How far in advance should power and cooling upgrades be ordered for Rubin racks?
Order upgrades immediately after the OEM power and thermal summaries are received. Long-lead items such as utility transformers require early coordination. Starting early prevents the rack delivery date from slipping.
Can existing air-cooled PDUs be reused with liquid-cooled racks?
Only if the OEM power summary confirms the existing PDU rating meets or exceeds the new rack draw and the connectors match. Most operators replace PDUs because next-gen racks require higher amperage or different outlet configurations.
What is the minimum aisle width needed for service access?
TIA-942-B specifies clearance on each side of the rack for liquid cooling service. This dimension must be maintained after containment and cable trays are installed.
Who performs the integrated load-bank test before racks arrive?
The operator or their designated commissioning agent runs the test. Leviathan Systems can witness and document results but does not own the facility power and cooling systems.
Do MPO trunks need to be tested before or after rack placement?
Test before racks are placed. Once racks occupy the row, access to pathways is restricted and any failed fiber requires partial de-installation to repair.