A standard CRPS power supply is compatible with a server only when “standard” resolves to a named specification, revision, form-factor implementation, selected options, and disclosed deviations—and those requirements match the actual chassis, power distribution board (PDB), and control architecture. CRPS is a standards family and an industry label, not a universal interchangeable-part system.
For an existing server, platform approval of the exact PSU part number, hardware revision, firmware, and airflow option is the strongest substitution evidence. Without that approval, buyers need a controlled comparison of the mechanical drawing, connector and pin map, condition-specific output capability, AC input range, management behavior, pairing policy, and failure-state performance. Similar wattage or a physically mating connector cannot close those engineering gaps.
Quick Answer
Treat a quotation for a standard CRPS power supply as incomplete until it identifies the governing base specification, revision, mechanical profile, options, and deviations. Match that declaration against the chassis bay and blind-mate interface, every PDB contact, required output under actual input and thermal conditions, airflow direction, BMC management requirements, and redundancy policy. Then qualify startup, load sharing, PSU loss, AC-feed loss, fault reporting, and permitted live replacement. “CRPS compatible” is useful only as a product-family description unless the supplier provides this evidence or names the exact qualified server platform.
What “Standard” Can—and Cannot—Mean in a CRPS Quotation
CRPS usually describes a removable server power module intended for a common redundant power architecture. The commercial term is used at several evidence levels:
- Standards-family construction: the PSU resembles a CRPS implementation but carries no complete conformance declaration.
- Mechanical-profile conformance: the housing and blind-mate location follow a documented form-factor profile, while electrical or management options may differ.
- Specification conformance: the manufacturer declares compliance with a named base document and applicable companion documents, including options and exceptions.
- Interoperability validation: the supplier or system integrator has tested the candidate with a defined PDB, BMC, and chassis configuration.
- Platform qualification: the server manufacturer has approved the exact PSU identity and revision for that platform.
These descriptions are not equivalent. Specification conformance controls some interfaces, but only platform qualification directly answers whether a particular replacement is approved for a particular server. A vendor-defined interoperability report can also be useful, provided it names the PSU configuration, chassis, PDB, BMC firmware, test conditions, and pass criteria.
A quotation such as “standard CRPS, PMBus, hot-swap, suitable for common server backplanes” leaves the key engineering questions unanswered. It does not identify which CRPS document applies, whether a management feature is mandatory or optional, which contacts are populated, how the output changes with input conditions, or which backplane was tested.
M-CRPS Is an Example, Not the Definition of Every CRPS Product
M-CRPS provides a useful example of why a broad CRPS label is insufficient. M-CRPS is an illustrative named implementation and must not be treated as the universal definition of every specification, legacy design, or vendor product marketed simply as CRPS. A product described as CRPS may follow another documented implementation, a previous industry interface, or a vendor-specific derivative. Equivalence must never be inferred from the acronym alone.
The current Open Compute Project DC-MHS specifications page lists the M-CRPS Base Specification and its associated mechanical drawings, with the applicable version and release date shown on that official page. Because OCP revises this specification set over time, buyers should use the currently listed controlled revision and linked drawings rather than relying on a copied historical filename or revision label. Publication by OCP does not certify any individual PSU.
The base specification and its associated mechanical materials have different scopes:
- The the applicable M-CRPS Base Specification is the source to inspect for applicable electrical behavior and interfaces. Relevant source locators are the clauses titled for output requirements, current sharing, power-supply timing, protection or fault behavior, control signals, and PMBus communication. Those clauses must be read with their conditions and optional-feature language.
- The the associated OCP mechanical drawings and implementation materials supply the controlled mechanical definition. The controlled mechanical drawings and their datum, envelope, connector-location, handle, latch, and keep-out callouts—not the phrase “185mm” by itself—are the authoritative locators for physical comparison.
Those subject locators should be converted into the exact clause, table, figure, drawing-sheet, and revision identifiers from the controlled PDF in the purchase specification. That prevents a copied clause number from an obsolete draft or reformatted document from becoming the contractual reference.
| Claim being evaluated | Controlled source locator | What the locator can establish | What still requires product or platform evidence |
|---|---|---|---|
| Mechanical profile | The currently listed M-CRPS mechanical drawings and implementation materials; controlled outline, chassis-interface, handle/latch, PDB-interface, and keep-out drawings | Drawing-defined datums, envelope relationships, insertion geometry, service features, and connector-position relationship for that implementation | Candidate tolerances, bay stack-up, guide engagement, insertion force, connector alignment, extraction clearance, and installed thermal behavior |
| Current sharing | The applicable M-CRPS Base Specification; clause titled “Current Sharing,” including its load conditions, tolerances, and applicable options | The behavior required by that clause when it applies to the declared implementation | Candidate conformance, approved mixed-unit policy, PDB current paths, firmware interaction, and sharing under the server’s real load profile |
| Management support | Base specification PMBus and status/communication clauses, plus every normative document they reference | Required or optional interface behavior within the selected profile | Implemented commands, addresses, scaling, inventory fields, alarm interpretation, polling behavior, and BMC interoperability |
| Startup and contact sequencing | Base specification timing/control clauses and companion form-factor connector drawings showing contact relationships | Applicable timing requirements and drawing-controlled mating relationships | PDB sequencing, chassis tolerances, BMC policy, startup under configured load, and behavior during insertion or removal |
| Fault response | Base specification protection, warning, shutdown, isolation, and status clauses applicable to the selected implementation | Specified response under the defined test conditions | Actual thresholds, recovery policy, event logging, PDB isolation, workload effect, and behavior of the surviving PSU |
This mapping is deliberately conditional. Current sharing, timing, management, sequencing, and fault behavior must not be described as universally controlled in the same way across every CRPS-labeled product. The governing revision, applicability language, selected options, and supplier deviations determine which requirements actually apply.
Build Compatibility from Three Evidence Layers
A defensible substitution decision uses three distinct evidence layers. Combining them into a single “compliant” checkbox obscures where risk remains.
1. Specification applicability
First identify the exact documents claimed by the manufacturer. Record the title, revision, mechanical profile, referenced base document, selected options, and exclusions. If a requirement is conditional, the supplier must state whether that condition applies to the candidate.
The conformance matrix should map each material claim to an exact clause, table, or drawing from the controlled source. It should also distinguish “conforms,” “not applicable,” “optional and implemented,” “optional and not implemented,” and “deviates.” A blank cell or generic check mark is not enough for an electrical interface used in an energized system.
2. Candidate PSU declaration
The product record must then identify what the actual module delivers. Relevant evidence includes:
- controlled mechanical outline and tolerance drawing;
- complete connector-contact map and signal definitions;
- rated output and input-dependent output limits;
- supported AC input and environmental conditions;
- airflow direction and permitted orientation;
- hardware, firmware, and management implementation;
- protection and recovery behavior;
- approved pairing or mixed-module restrictions.
These attributes cannot safely be inherited from another PSU in the same marketing family. A different firmware build, fan orientation, connector population, or internal configuration can affect platform behavior even when the outer shell is unchanged.
3. Complete-platform qualification
The final layer tests the candidate in the intended chassis with the applicable PDB, BMC firmware, rack input, and workload. Qualification should cover normal startup and shutdown, steady load, workload transitions, current sharing, single-PSU loss, single-feed loss, fault reporting, recovery, and any authorized live-service procedure.
The compatibility decision can be represented as an intersection:
Applicable specification requirements ∩ candidate declaration ∩ chassis/PDB requirements ∩ validated platform behavior
A missing layer leaves an unresolved assumption rather than a compatible system.
Mechanical Verification Continues Through the Blind-Mate Interface
Mechanical verification does not end when the PSU enters the bay; it continues through full engagement of the blind-mate connector and the service mechanism. Drawings should be compared from common datums rather than from exterior dimensions or nominal length.
The review needs to account for:
- bay opening and insertion path;
- guide-rail engagement and allowable movement;
- latch location, retention, and handle travel;
- PSU and PDB connector position from controlled datums;
- engagement depth and tolerance stack-up;
- rear and side keep-out regions;
- extraction clearance and technician access;
- possible interference from cables, covers, or nearby components.
A module can enter the bay yet stop before complete contact engagement. It can also load the PDB connector laterally if the chassis guides, PSU rails, or connector mounts do not share the expected geometry. Either condition can cause intermittent operation, heating, damaged contacts, or failure to preserve the intended mating sequence.
The “185mm” designation identifies a documented profile; it is not permission to verify only nominal depth. Acceptance should be based on the complete controlled drawing and the chassis stack-up.
The Pin Map Is the Electrical Compatibility Contract
Once physical alignment is established, every PSU contact must be reconciled with the PDB net map. The engineering record should show, for each populated position:
- PSU-side function and PDB-side destination;
- power, return, standby, signal, or management classification;
- direction and asserted state for control signals;
- electrical levels or definitions from the applicable documents;
- current allocation for power contacts where documented;
- mating sequence or contact-length relationship where applicable;
- required response if the contact is open, shorted, or asserted.
A connector shell can mate while one implementation assigns a different function to a signal position, leaves an expected contact absent, or uses different logic. An energized trial is not an acceptable way to discover that mismatch.
The PDB may also impose a lower usable output than the PSU nameplate. Connector contacts, copper planes, current-sense elements, isolation devices, protective components, and cooling determine how much current the path can carry. A higher-rated standard CRPS power supply does not upgrade those components.
Redundant Capacity Is a Single-Survivor Requirement
Rated output is a capability under stated conditions, not a measurement of server consumption. Likewise, two installed PSUs do not necessarily provide twice one module’s rating as usable redundant capacity.
In a 1+1 design, two active modules may share the normal load. After one PSU or one AC feed fails, the remaining module and its PDB path must support the required server load. The relevant continuous ceiling is:
Surviving-path capability = min(PSU capability under deployment conditions, PDB path limit, platform-approved limit)
The PSU term must reflect the actual AC input, inlet temperature, altitude, airflow, and any manufacturer-specified derating. The PDB and platform limits must come from controlled server documentation or qualification—not from the replacement PSU’s label.
Hypothetical selection calculation
Consider an anonymized evaluation with explicit illustrative assumptions:
- candidate PSU nominal rating: 1,600 W;
- documented output at the planned input and thermal condition: 1,450 W;
- validated single-path PDB limit: 1,400 W;
- required maximum post-failure server load: 1,250 W.
The surviving-path capability would be:
min(1,450 W, 1,400 W) = 1,400 W
Continuous headroom would be:
1,400 W − 1,250 W = 150 W
That result does not by itself prove adequate margin. The platform owner must define the required measurement allowance and operational reserve. A documented capacity requirement might be constructed as:
Required capacity = validated maximum continuous load + measurement uncertainty + approved reserve
If those illustrative components were 1,190 W, 35 W, and 50 W, respectively, the requirement would be 1,275 W. The 1,400 W path would then have 125 W of documented continuous margin. These example values are not CRPS requirements or universal sizing percentages.
Transient behavior must be evaluated separately. An accelerator or CPU excursion has a magnitude, duration, repetition rate, and interaction with protection thresholds. The qualification plan should define the workload transient, acceptable voltage response, PSU recovery, PDB behavior, and pass criteria. Continuous headroom cannot be assumed to prove transient support.
If both PSUs initially share a 1,250 W load, each may carry approximately 625 W. Following a failure:
Post-failure load per surviving PSU = 1,250 W ÷ 1 = 1,250 W
The system may contain 3,200 W of installed nameplate capacity, yet its illustrated condition-specific redundant path is only 1,400 W before applying required reserves and transient criteria.
Input Power Must Be Checked in the Surviving-Feed State
Compatibility also depends on the AC source. A candidate may have different output capability or input current at different line conditions. Buyers should verify the model-level input range, frequency, current, output derating, power-factor data, and environmental conditions.
A first-order conversion is:
AC input power ≈ DC output ÷ efficiency at that operating point
If a hypothetical survivor provides 1,250 W DC and 94% efficiency is assumed only to illustrate the equation:
1,250 W ÷ 0.94 ≈ 1,330 W AC
The assumed efficiency is not a product claim and should not be used for final circuit design. Deployment planning needs the candidate manufacturer’s input-current data and applicable site electrical requirements.
Normal operation can hide the limiting state. With two independent feeds and approximately equal load sharing, each feed sees only part of the server demand. Following feed loss, the surviving PDU branch may carry the entire server input. That state must fit the permitted branch, receptacle, PDU, and rack capacity while preserving the site’s required operating margin.
Management and Live Service Need Separate Approval
“PMBus supported” identifies a protocol family, not complete BMC interoperability. The candidate can differ in supported commands, device addressing, telemetry scaling, status-bit meanings, inventory fields, warning thresholds, optional controls, and firmware behavior.
A useful management matrix lists each function required by the server and records whether the candidate supports it, how the BMC uses it, and how it was tested. At minimum, qualification should distinguish:
- Power-path behavior: startup, regulation, sharing, isolation, shutdown, and recovery.
- Monitoring: identity, voltage, current, temperature, warnings, faults, and redundancy state.
- Platform response: event logging, alerts, fan response, power policy, and workload action.
A successful boot validates none of the management details by itself. The server might remain powered while reporting false redundancy, incorrect telemetry, or no warning for a developing fault.
Hot-swap also requires platform-level approval. Removability does not prove that energized extraction and insertion are safe in every bay. The PSU interface, connector sequencing, PDB isolation, BMC behavior, surviving capacity, and written service procedure must all support the operation. If the remaining module cannot carry the required load, a physically hot-swappable interface does not make removal operationally safe.
Airflow Is a Configuration Attribute
The mechanical profile does not establish airflow direction or thermal suitability. The candidate record should identify intake and exhaust orientation, while the platform documentation defines the intended server flow path.
An airflow mismatch can recirculate exhaust or oppose the chassis fans. It may pass a short, low-load bench test but fail during sustained load, elevated inlet temperature, or single-PSU operation. Thermal qualification should therefore include the intended workload and the relevant failure state, with the survivor carrying the full required load.
A larger nameplate rating does not guarantee lower temperature. Conversion losses at the actual operating point, fan-control policy, impedance, current sharing, and internal protection thresholds all affect thermal behavior.
Final Procurement Decision Gate
Approve a standard CRPS power supply only when the procurement file answers all of the following with controlled evidence:
- Named specification: What base specification and companion documents does the supplier claim?
- Revision and options: Which revisions, profiles, conditional features, and deviations apply?
- Mechanical drawing: Do common-datum drawings confirm bay, guide, latch, handle, connector, keep-out, and extraction compatibility?
- Pin map: Does every populated PSU contact agree with the PDB net, electrical definition, and mating sequence?
- Power envelope: Can one PSU and one PDB path support the required continuous and transient load under deployment conditions?
- Input range: Is the required output available on the actual AC source, and can each feed support the failure state?
- Airflow: Does the airflow option match the chassis thermal architecture?
- Management matrix: Do the implemented commands, telemetry, status, alerts, and BMC responses match platform requirements?
- Pairing policy: Are mixed models, ratings, hardware revisions, firmware versions, or airflow options permitted?
- Qualification evidence: Has the exact configuration passed normal operation, startup, load transitions, PSU loss, feed loss, fault reporting, recovery, and any authorized live-service procedure?
The conditional answer is straightforward: a standard CRPS power supply is a defensible replacement when a named and revision-controlled standards implementation matches the product declaration and the complete server platform passes qualification. If the evidence establishes only a CRPS-family shape, nominal wattage, or broad compatibility statement, the module remains an unqualified candidate rather than an interchangeable server PSU.
