Liquid Cooling_
Coolant Chemistry & Maintenance for Direct Liquid Cooling
Specifies the chemistry, treatment, sampling, and maintenance steps for PG25-based direct liquid cooling loops on GPU racks, including when to choose PG25 over water and how to keep loops free of fouling and corrosion.
Key facts
- PG25 propylene glycol mixtures are selected for closed direct liquid cooling loops when freeze protection or additional lubricity is required by the rack OEM.
- Water-only loops demand stricter material compatibility and higher biocide dosing because they lack the inherent corrosion inhibition of glycol formulations.
- Biocide selection must match the specific elastomer and metal mix in the cold plates, manifolds, and hoses to avoid swelling or leaching.
- Side-stream filtration at the rating specified by the OEM is required to capture particulates before they reach GPU cold-plate channels.
- Fluid sampling for pH, conductivity, inhibitor concentration, and microbial counts follows the interval set by the OEM or trend analysis on production AI racks.
- Maintenance cadence is driven by the first of either OEM-specified fluid life or measured deviation outside control limits rather than fixed calendar time.
- Failure to maintain inhibitor levels allows galvanic corrosion between dissimilar metals in the loop within months of commissioning.
PG25 versus water in GPU rack loops
PG25 is chosen when the rack design or site conditions require freeze protection or when the OEM specifies glycol for seal lubrication and reduced galvanic activity. The concentration follows the rack OEM data sheet to balance heat capacity against corrosion protection and viscosity. Water-only loops are used only when the entire bill of materials has been validated for de-ionized water service and when the facility maintains strict air-side humidity control to eliminate condensation risk.
Field crews verify the as-installed fluid against the rack OEM data sheet before first power-on. Mixing PG25 into a water loop or vice versa changes both freeze point and inhibitor effectiveness, so any top-off must come from the same formulation drum. Leviathan Systems records the batch number and refractive index reading at commissioning to establish the baseline for later sampling.
Biocide selection and dosing
Biocide chemistry must be compatible with the specific EPDM, FKM, and nickel-plated surfaces present in the cold plates and quick-disconnects. Incompatible products cause seal swell or strip protective films, accelerating corrosion. Dose rates follow the biocide manufacturer’s label once loop volume is known; over-dosing raises conductivity and promotes foaming.
Biocide is introduced during initial fill and again only after laboratory results show microbial activity above the control limit. Continuous low-level dosing via metering pump is avoided because it masks the root cause of contamination and complicates later disposal.
Filtration placement and rating
A side-stream filter sized to a portion of total loop flow per OEM guidance captures wear particles and any residual construction debris before they reach the narrow GPU cold-plate passages. The filter housing uses the same quick-disconnect standard as the rest of the rack so cartridges can be changed without draining the loop. Differential pressure across the filter is monitored; a sustained rise above baseline indicates the cartridge must be replaced.
Magnetite and copper fines are the most common captured materials once the loop has run for several months. Filter media is selected to avoid shedding fibers that could themselves become particulate downstream.
Fluid sampling protocol
Samples are drawn from a dedicated low-point valve after the circulation pump has run long enough to circulate the loop fully. The initial volume is discarded to flush stagnant fluid from the sample port. Subsequent aliquots are collected in clean, labeled bottles for pH, conductivity, reserve alkalinity, copper and iron ion counts, and total aerobic bacteria plate count.
Results are compared against the OEM control bands established at commissioning. Any parameter outside band triggers a work order before the next scheduled sample interval. Leviathan Systems archives the laboratory reports with rack serial numbers so trend data can be reviewed during annual audits.
Maintenance cadence and decision criteria
The loop is inspected visually and sampled at the interval set by the OEM during the first year, then adjusted once stable chemistry is confirmed. Fluid is replaced when any single measured parameter exceeds the OEM limit or when cumulative runtime reaches the fluid manufacturer’s stated life, whichever occurs first. Partial drains and refills are avoided because they dilute inhibitors unevenly.
Before any fluid change, the filter is swapped and the loop is flushed with a compatible cleaner if iron or copper levels indicate active corrosion. Post-maintenance verification includes refractive index, conductivity, and a final particle count before the rack is returned to production.
Common failure modes observed in the field
The most frequent issue is low inhibitor residual after several months because scheduled sampling was skipped or the laboratory report was not acted upon. This produces pitting on aluminum cold plates and copper manifold braze joints, visible as deposits at quick-disconnects. The second common failure is microbial fouling after a biocide that was incompatible with the installed hose material was used; the resulting slime restricts flow and raises GPU delta-T.
Construction debris left in manifolds because the initial flush used insufficient velocity also appears regularly. These particles lodge in cold-plate channels and create hot spots that are only discovered after the rack has been in service for weeks. Both modes are caught by trending conductivity and pressure drop rather than waiting for thermal alarms.
Standards referenced: OEM rack coolant specification · ASHRAE TC 9.9 liquid cooling guidelines · ASTM D1384 corrosion test method
Frequently asked_
How often should PG25 loops be sampled on a production GPU rack?
Sampling follows the interval set by the OEM for the first twelve months. After stable chemistry is confirmed, the interval may extend provided no parameter has approached its control limit. The decision is always driven by measured values rather than calendar time alone.
Can a water-only loop be converted to PG25 after installation?
Conversion requires complete draining, flushing, and verification that every elastomer and coating remains compatible with the new fluid. Most operators avoid the change because the required volume of PG25 and the disposal cost exceed the benefit once the rack is already commissioned on water.
What particle size filter is required upstream of GPU cold plates?
The side-stream filter uses the rating specified by the OEM. The filter must be located so that all flow to the cold plates passes through it, and differential pressure is trended to determine change-out timing.
Which laboratory tests are mandatory on every coolant sample?
pH, conductivity, reserve alkalinity or inhibitor concentration, copper and iron ion levels, and total aerobic bacteria count are required on every sample. Additional tests for glycol percentage or specific biocide residual are added only when the primary results indicate a problem.
Who performs the fluid analysis for Leviathan Systems deployments?
An accredited coolant laboratory receives the samples and returns results against the OEM control bands. Leviathan Systems maintains the chain-of-custody log and archives the reports with each rack serial number for the life of the installation.