A procurement team receives a proposal for two 1500W hot-swap modules in a 1+1 server power shelf. The workload estimate is 1,760W DC, so the spreadsheet shows “3,000W installed” and appears to leave ample capacity. That interpretation fails the first outage test: after either module is removed or faults, the remaining unit would be asked to deliver 1,760W. A 1500w 80 plus titanium server psu can contribute 1,500W of rated DC output under its approved operating conditions, but installing two does not create 3,000W of failover capacity.
The more useful procurement question is how each module moves among 150W, 300W, 750W, and 1,500W output levels. Those are the 10%, 20%, 50%, and 100% certification load points for a 1,500W unit. In an equal-sharing 1+1 system, the module may sit at one load point during normal operation and jump to another when its partner fails. That movement determines which Titanium threshold is relevant, where the certificate says nothing exact, and whether the surviving module remains within its rated envelope.
Four module load points become different system load points
Efficiency certification applies to the load on the individual power supply, not automatically to the total output of the server or power shelf. For a 1,500W module, the arithmetic is straightforward:
- 10% module load is 150W DC.
- 20% module load is 300W DC.
- 50% module load is 750W DC.
- 100% module load is 1,500W DC.
With two modules sharing equally, each certification point corresponds to twice as much aggregate DC demand during normal operation. A total 300W server load places approximately 150W on each module. A total 600W load places 300W on each. A total 1,500W load places 750W on each. A total 3,000W load would place 1,500W on each, but it would consume all installed output capacity and leave no capacity for the loss of a module.
After one module fails, the same aggregate demand transfers to the survivor. The system load has not doubled; the output assigned to one module has. This creates several useful load-point transitions.
| Illustrative aggregate DC demand | Per module with equal sharing | Normal module load | Surviving module after failure | Failover result |
|---|---|---|---|---|
| 180W | 90W | 6% | 180W, or 12% | Within rating, but normal operation is below the 10% certification point |
| 300W | 150W | 10% | 300W, or 20% | Moves from the 10% to the 20% certification point |
| 600W | 300W | 20% | 600W, or 40% | Starts at a certification point; lands between specified points |
| 750W | 375W | 25% | 750W, or 50% | Lands on the 50% certification point |
| 1,500W | 750W | 50% | 1,500W, or 100% | Moves from the 50% point to full rated output |
| 1,760W | 880W | About 58.7% | 1,760W, or about 117.3% | Exceeds a 1,500W survivor’s continuous nameplate rating |
These figures assume equal sharing and constant aggregate demand solely to illustrate the sizing logic. Actual sharing can deviate between modules, and system power can change during a fault because fans accelerate, accelerators change state, drives spin up, or platform controls shed load. Procurement should therefore use the highest credible post-failure DC requirement rather than treating the equal-share result as a complete engineering model.
The certificate category must match the redundant application
For the CLEAResult 80 PLUS program, the relevant Titanium criteria depend on the product category. According to the official 80 PLUS PSU ratings explanation, 230V Internal Redundant Titanium criteria are 90%, 94%, 96%, and 91% efficiency at 10%, 20%, 50%, and 100% load, respectively.
The 230V EU Internal Non-Redundant Titanium criteria are different: 90%, 93%, 95%, and 91% at those same load percentages. The difference at 20% and 50% matters when a procurement specification quotes “Titanium” without identifying the category. A hot-swap appearance, CRPS description, or redundant server application does not by itself prove that a particular module has a 230V Internal Redundant Titanium certificate. The exact model must be matched to its official certification record.
For a correctly identified 1,500W 230V Internal Redundant Titanium certificate, the formal test points translate to the following minimum efficiencies:
- At 150W DC output, the 10% criterion is 90%.
- At 300W DC output, the 20% criterion is 94%.
- At 750W DC output, the 50% criterion is 96%.
- At 1,500W DC output, the 100% criterion is 91%.
Those percentages are certification test minima at defined conditions. They are not a promise that every installed module will produce exactly those results at every input voltage, temperature, airflow condition, manufacturing tolerance, or intermediate load. Nor does a threshold specify exact efficiency at 6%, 12%, 25%, 40%, 60%, or 80% load. A curve, report, or manufacturer data specific to the exact unit is needed when those operating regions materially affect the energy model.
The load-point sequence also explains why the highest published Titanium percentage should not be applied to all operating states. The 96% criterion belongs to the 50% point in the 230V Internal Redundant category. In the 1,500W failover example, each module may operate around 750W and therefore near that formal point while both are healthy. Following a failure, the survivor moves to 1,500W, where the certification criterion is 91%, not 96%.
DC output ratings and AC input calculations belong on opposite sides of the PSU
The 1,500W nameplate is a DC output rating. It describes power available to the server-side rails within the manufacturer’s approved operating envelope. It does not mean the module draws only 1,500W from the facility. AC input must cover DC output plus conversion loss.
Efficiency is calculated as DC output divided by AC input. At a stated efficiency, the corresponding input is:
AC input = DC output ÷ efficiency.
Using the 230V Internal Redundant Titanium minimum thresholds only as mathematical certification boundaries, 150W DC at 90% corresponds to about 166.7W AC input. At 300W and 94%, the result is about 319.1W AC. At 750W and 96%, it is about 781.3W AC. At 1,500W and 91%, it is about 1,648.4W AC.
These are not predictions for an unidentified PSU in a deployed server. They show why facility power cannot be obtained by adding DC nameplate ratings, and why using the peak 96% value at full output understates input demand. At the certification thresholds, the implied conversion loss rises from about 31.3W at 750W output to about 148.4W at 1,500W output. Full-load failover can therefore change both the module loading and the upstream AC demand substantially.
Upstream branch circuits, rack power distribution, and cooling calculations require more than this threshold arithmetic. They should account for the exact PSU data, actual input voltage, power factor information, environmental limits, server utilization behavior, and any applicable facility design margins. The 230V certification category also should not be used as proof of identical efficiency or full output capability across every supported input condition.
Idle sharing can fall below the certified map
Consider an assumed storage or telecom server that consumes 180W DC during a deep idle period. If two modules share equally, each supplies approximately 90W, or 6% of its 1,500W rating. The 230V Internal Redundant Titanium criteria begin at 10% load, so the certificate does not establish a numerical minimum at that 6% operating point.
If one module fails while the demand remains 180W, the survivor moves to 12% load. That position is above the 10% test point but below the 20% test point. It would still be incorrect to interpolate a guaranteed efficiency between 90% and 94%. Certification provides minima at specified points, not a guaranteed straight line connecting them.
Some platforms may use power-management behavior that changes how redundant modules share low loads. One module could carry more of the demand while another remains in a different operating state, but that behavior is platform- and PSU-dependent. It should not be assumed from the module wattage or hot-swap format. Module telemetry, platform documentation, and an approved power-management policy are better evidence of the real idle distribution.
This distinction changes the procurement analysis. A server that spends most of its life at 180W may have normal module operation below the certificate’s lowest formal load point, even though a failure moves the survivor into the certified range. By contrast, a server drawing 1,500W normally places both modules at the 50% point and then pushes the survivor to the 100% point. The same pair of 1,500W supplies therefore presents two very different efficiency and thermal operating patterns.
The single-module boundary must include real failover behavior
A valid 1+1 sizing exercise starts with the output demanded after one complete power path is unavailable. For two 1,500W modules, the basic continuous condition is:
Required post-failure DC output must remain within the approved capability of one module under the actual operating conditions.
The calculation should include loads powered through the PSU rather than relying only on processor or accelerator thermal design power. Depending on the server architecture, the total can include CPUs, GPUs or other accelerators, memory, storage, network adapters, fans, pumps, motherboard conversion stages, and peripheral loads. Losses in downstream voltage regulators are already part of what the PSU must supply if measurements are taken at its DC output.
The relevant demand is also not necessarily the average shown by a workload dashboard. A failure can coincide with a compute burst, fan response, battery charging state, storage activity, or restart event. The PSU and platform may define rules for peak duration, transient response, temperature, airflow, input voltage, or current sharing. Those manufacturer-approved limits govern whether a peak is acceptable. An undocumented assumption that a module can continuously exceed 1,500W should not be used to close a capacity gap.
A worked conditional procurement case
Assume, for illustration, that a proposed AI server has a measured steady DC output demand of 1,320W, an engineering estimate of 1,440W under its highest validated workload, and a short platform event that can reach 1,520W. Also assume that no approved load-shedding policy has yet been documented.
With both modules healthy and sharing equally, the expected per-module outputs would be 660W, 720W, and 760W. Those values are about 44%, 48%, and 51% of rating. They cluster around the 750W certification point, but only 750W is the exact 50% point. The certificate does not assign a precise installed-system efficiency to 660W, 720W, or 760W.
After a module failure, the survivor would face the full 1,320W steady demand and the 1,440W validated maximum. Both are below the 1,500W DC nameplate, subject to the exact input, thermal, airflow, and platform limits. The assumed 1,520W event crosses the continuous rating. Whether that event is supportable cannot be decided from “Titanium,” from the presence of two modules, or from simple average-load arithmetic. It requires an approved transient specification for the exact PSU-platform combination, or a control that keeps post-failure demand at or below the accepted boundary.
If the manufacturer permits a defined transient above nameplate under the relevant conditions, that documented envelope may resolve the event. If not, the purchaser needs a lower failover workload, enforced power capping or load shedding, or a power architecture with greater single-fault capacity. Normal sharing does not cure the 1,520W survivor requirement.
Installed capacity and redundant capacity should be recorded separately
For a two-module 1+1 arrangement, procurement records benefit from two distinct values:
- Installed nominal output: up to 3,000W across two 1,500W modules when both are available, subject to platform limits.
- Single-module continuous boundary: up to 1,500W from the survivor under the exact approved operating envelope.
The first value describes total nameplate capacity. The second determines whether full workload can continue after one module or power path is lost. If normal server demand is above 1,500W, the pair may still operate while both modules are healthy, but it is no longer providing full-load 1+1 continuity unless the platform automatically reduces demand to an approved level.
More modules can support a larger redundant design, but only when the chassis, power shelf, distribution topology, firmware, and manufacturer documentation define that architecture. Arithmetic such as “three times 1,500W” does not by itself establish a 2+1 system, and a wattage label does not prove that modules from different families are physically or electrically interchangeable. CRPS mechanics, connectors, output behavior, communications, airflow, firmware expectations, and platform qualification all remain part of the selection.
The purchase line should state the operating transition
A useful sourcing requirement identifies the exact PSU model, its official 230V Internal Redundant Titanium record if that certification is required, the server or power-shelf approval, supported input and environmental conditions, and the maximum allowed DC demand after one module is lost. It should also state whether that post-failure figure is sustained by hardware capacity or by a validated power cap.
If the validated single-failure demand remains at or below 1,500W throughout the approved envelope, two qualified 1,500W modules can be considered for full-load 1+1 operation, with normal and failed-state efficiency evaluated at their respective module loads. If demand can reach 1,760W and no approved reduction occurs on failure, the purchase should not be described as 1+1 at that workload, regardless of the 3,000W installed nameplate total or the Titanium label.
