Compatibility is deceptive.
When a buyer sees “CRPS compatible” beside a server chassis and “CRPS” beside a 2,400 W power supply, it is tempting to treat those two labels as a mechanical and electrical guarantee, even though width, depth, connector position, PDB architecture, input-voltage derating, firmware communication, airflow direction, latch geometry, and redundancy design can still disagree.
So what exactly have you proved?
Not much.
That is the uncomfortable answer.
CRPS—Common Redundant Power Supply—exists specifically to reduce the chaos that once surrounded proprietary redundant server power supplies. The Open Compute Project’s DC-MHS work now includes M-CRPS specifications intended to create consistent power-supply interfaces for modular server platforms. OCP’s published documentation includes the M-CRPS 1.05 RC5 specification submitted on October 1, 2024, together with an October 2024 M-CRPS design-specification release package.
Standardization helps enormously.
But standardized does not mean universal.
And that distinction is where expensive chassis mistakes begin.
I use a stricter definition than the one I often see on purchasing spreadsheets.
A server chassis is genuinely compatible with a CRPS power supply only when the entire installed power subsystem works together:
Miss one and the chassis can fail even though both product descriptions contain the letters “CRPS.”
This is why I would never approve a custom server enclosure from a line in an RFQ that says simply:
Power Supply: 2 × CRPS
That is not an engineering specification.
It is a clue.
Before chassis CAD is frozen, the PSU should be treated like the motherboard, GPU, backplane, fan wall, and drive cage: a defined mechanical and electrical component. The same principle appears throughout our guidance on the key design inputs for a custom chassis project—design around the final hardware architecture rather than vague component categories.
Here is where the assumption becomes dangerous.
Many widely used M-CRPS supplies follow a compact envelope around 40 mm high × 73.5 mm wide × 185 mm deep. Advanced Energy, for example, currently lists its 2.4 kW CSU2400AT M-CRPS supply at 1U × 73.5 × 185 mm, with up to 196.7 A output and peak efficiency of 96%.
Easy, right?
No.
HPE’s current Modular Common Redundant Power Supply documentation lists two M-CRPS widths: 73.5 mm and 60 mm. HPE also explicitly states that its new M-CRPS design is not compatible with servers prior to Gen12.
Read that again.
Two products can legitimately be described in the market as M-CRPS while still demanding different mechanical provisions at chassis level.
That single fact destroys the lazy buying rule:
“CRPS is CRPS.”
It isn’t.
Do not request only PSU length, width, and height.
Request:
For a standard server chassis, millimeters matter.
For a 1U system, they matter even more.
If chassis height is still being selected, review the thermal and mechanical tradeoffs in 1U vs 2U vs 4U server chassis form factors before locking the PSU architecture. A power subsystem that is easy to package in 4U may become ugly very quickly in 1U.
A CRPS PSU usually does not behave like a consumer ATX power supply with a bundle of motherboard connectors hanging from it.
The server architecture commonly places a Power Distribution Board, or PDB, between the redundant PSU modules and the downstream motherboard, storage, GPU, fan, or auxiliary power loads.
That board matters.
A lot.
The CRPS modules may provide high-current 12 V power and standby power through a card-edge interface. The PDB then becomes responsible for distributing that power through the system and, depending on the architecture, providing additional conversion, protection, sensing, management, or output connectors.
So when a supplier tells me:
“Yes, our chassis accepts CRPS.”
My next question is:
Which PDB?
If the answer is vague, the compatibility review is unfinished.
Ask for:
This is where I become unpopular with vague quotations.
Good.
A $30 metal-panel change is inconvenient.
A power-board redesign after chassis tooling, prototype approval, cable production, and motherboard integration is much worse.
| Compatibility Item | What Must Match | Typical Failure |
|---|---|---|
| PSU width | Chassis cage and guide width | 60 mm PSU specified for a 73.5 mm mechanical design |
| PSU depth | Internal envelope and connector location | PSU fits opening but cannot fully engage |
| Hauteur | Chassis vertical clearance | Interference in dense 1U packaging |
| Output connector | PSU and PDB mating interface | Physical insertion without electrical compatibility |
| PDB rating | System current and peak load | Connector or copper path overheats |
| PSU quantity | 1+1, 2+1, N+1, N+N architecture | Redundancy disappears at full load |
| AC input | Site voltage versus rated output | “2,000 W” PSU provides much less power at low line |
| Main output | System distribution architecture | Incorrect voltage or downstream conversion |
| Standby supply | Motherboard/platform requirement | Management functions fail when system is off |
| PMBus/SMBus | Platform management support | PSU health or telemetry unavailable |
| Débit d'air | Chassis front-to-rear cooling path | PSU fights system airflow |
| Hot-swap clearance | Rack and rear service area | PSU technically removable only after disconnecting other hardware |
| Power cord inlet | C14/C20 or other inlet and cable | Wrong rack PDU or cable selected |
| Efficiency/compliance | Deployment market requirements | Finished server misses procurement or regulatory target |
I would attach a version of this table to every serious chassis RFQ.
For custom projects, our complete chassis RFQ guide provides a broader framework for locking drawings, components, tolerances, quantities, inspection requirements, and production assumptions before quotations are compared.
This one catches buyers because the product name itself feels authoritative.
Imagine the BOM says:
2 × 2,000 W CRPS PSU
Someone enters 4,000 W into a spreadsheet.
Done.
Except the rack operates on a voltage at which the selected PSU cannot provide its full advertised output.
A current Lite-On 2,000 W CRPS example illustrates the problem beautifully. Its published data lists:
The same PSU also lists a 73.5 × 40 × 185 mm enclosure and a 12.2 V high-line output reaching roughly 163.5 A.
This is not a minor specification detail.
At low-line input, the nominal “2,000 W” unit in this example becomes a 1,000 W source.
Half.
Now imagine a GPU server that was sized under the assumption that each PSU would deliver 2 kW regardless of facility input.
Would you want to discover that after the machines arrive?
The RFQ should state:
Then request the PSU’s rated output at that exact input range.
Do not use only the model number’s headline wattage.
And do not confuse aggregate PSU nameplate power with fault-tolerant system power.
This misunderstanding refuses to die.
Two 2,000 W power supplies installed side by side do not automatically mean a 4,000 W fault-tolerant server.
If the system is designed as 1+1 redundancy, either supply should be capable of carrying the required system load after the other PSU fails or is removed.
That changes the math completely.
Suppose:
In a properly engineered 1+1 design, losing one PSU still leaves enough capacity.
Now suppose the server can demand 2,700 W.
Both supplies may carry that load together under normal conditions, but one 2,000 W module cannot maintain it after a failure.
You have parallel power.
You do not have full 1+1 redundancy at 2,700 W.
That distinction belongs in the purchasing specification.
I care more about this number than the giant wattage printed on a product page:
Maximum supported system load with one PSU unavailable.
That is the number that tells me whether the claimed redundancy actually survives the failure scenario the buyer thinks it survives.
For GPU and AI servers, calculate:
CPU + GPU + memory + storage + fans + motherboard + PCIe cards + pumps + PDB/DC-DC losses + transient margin
Then repeat the calculation under the intended degraded PSU state.
This discussion is becoming more important, not less.
The U.S. Department of Energy’s Lawrence Berkeley National Laboratory published a 2025 update in June 2026 estimating that data centers could account for 11.8% of total U.S. electricity use by 2030, with modeled scenarios ranging from 9.5% to 15.3%.
The earlier DOE-backed 2024 report estimated 176 TWh of U.S. data-center electricity consumption in 2023 and projected 325–580 TWh by 2028.
That macro number matters at chassis level because dense AI and accelerated-computing systems keep pushing more current through smaller mechanical spaces.
The metal box is not passive anymore.
Its layout determines:
For high-power accelerator projects, I would review the power subsystem together with the available GPU and AI server chassis architectures rather than treating the PSU as something added after the enclosure is chosen.
Another habit I dislike: choosing a CRPS PSU entirely by peak wattage.
Efficiency matters because every watt lost in the PSU becomes heat that the cooling system has to remove.
There is also a compliance angle.
For covered servers and online data-storage products placed on the EU market, Commission Regulation (EU) 2019/424 establishes minimum PSU efficiency requirements. From January 1, 2023, its table for single-output PSUs specifies minimum efficiencies of 90% at 10% load, 94% at 20%, 96% at 50%, and 91% at 100%, together with a minimum 0.95 power factor at 50% load. Scope and exemptions still need to be checked for the finished product.
That 50%-load figure is interesting.
Current high-end supplies can operate around that territory: Advanced Energy’s 2.4 kW M-CRPS product advertises peak efficiency of up to 96%.
But do not buy the badge and stop thinking.
Ask for the efficiency curve at your expected operating load.
A 3,200 W PSU running a lightly loaded appliance can make less sense than a smaller supply operating closer to its efficient zone, unless the larger unit is required for redundancy, peaks, future expansion, or platform standardization.
The PSU fan is part of the server’s airflow system.
Not decoration.
If the chassis is designed for front-to-back airflow but the selected PSU moves air in the opposite direction, several ugly things can happen:
And because the PSU often occupies a corner of the chassis, the problem can remain local enough that a CPU temperature test does not immediately expose it.
This is why power and thermal design belong in the same engineering review.
Our guide to designing front-to-back server chassis airflow explains how fan pressure, bypass air, drive cages, backplanes, cables, GPUs, and rear restrictions interact. Add the CRPS PSU to that same airflow map.
Literally.
Get the PSU drawing.
Find the airflow arrow.
Place it into the chassis CAD.
Then ask where the hot air goes.
This takes minutes.
Fixing the opposite condition after metal production takes longer.
Mechanical compatibility gets the attention because it is visible.
Management compatibility often gets ignored until integration.
Many modern CRPS and M-CRPS platforms use PMBus/SMBus communication for functions such as:
Advanced Energy’s current 2.4 kW M-CRPS, for example, includes digital control over an I²C interface and PMBus support.
A PSU can therefore deliver 12 V successfully while the BMC still complains.
That is a compatibility failure too.
Maybe not on a bench.
Definitely in a managed server fleet.
So ask whether the motherboard/BMC expects specific PSU addressing, telemetry, firmware behavior, or platform identification.
I have a simple test for “hot-swappable.”
Can a technician actually remove the failed CRPS module from the installed server without moving the rack, disconnecting unrelated cables, removing a PDU, or fighting a cable-management arm?
If not, I would hesitate to celebrate the feature.
Server serviceability has geometry.
Rear space may contain:
Your CRPS module may need 185 mm of internal space but considerably more external space for extraction.
That is why chassis power-supply compatibility should be reviewed together with required rack-depth and service-clearance calculations.
A rack can be deep enough for the chassis and still make PSU service miserable.
Before approving a server chassis, I would want this package.
That last group saves projects.
A substitute PSU should not enter production merely because procurement found one with the same wattage.
“Supports CRPS PSU.”
I dislike that sentence unless it is followed by a model list or an interface drawing.
Because what does “supports” mean?
The opening is the right size?
A PSU slides in?
The PDB mates?
The BMC recognizes it?
Both PSUs share load?
The system survives removal of one module?
Full output is available at 208 Vac?
The PSU can physically be extracted inside the target rack?
Those are seven different questions.
A professional quotation should tell you enough to answer them.
If you are buying a standard chassis, request the manufacturer’s validated CRPS PSU and PDB compatibility list.
If you are developing a custom server chassis, send the exact PSU and PDB drawings before the first mechanical layout is released.
That is the difference between “compatible on paper” and compatible in production.
CRPS power supply compatibility means that the selected PSU, server chassis, power distribution board, electrical interface, input voltage, redundancy architecture, airflow direction, management signals, and service clearances operate together as one validated power subsystem rather than merely sharing a CRPS label or similar external dimensions.
A proper review should therefore confirm exact model numbers and drawings rather than accepting “CRPS supported” as sufficient evidence.
No, CRPS power supplies are not guaranteed to have one universal physical size, because current M-CRPS implementations can use different mechanical widths and platform-specific designs even when they follow related industry specifications; HPE, for example, currently documents both 60 mm and 73.5 mm M-CRPS widths.
Depth, latch location, connector datum, handle geometry, and chassis guide structure should still be checked against the exact PSU model.
No, a CRPS PSU cannot automatically be assumed to work in every chassis advertised as CRPS compatible, because the PSU and chassis must also agree on mechanical dimensions, card-edge connection, PDB interface, current capacity, standby power, communication signals, redundancy design, input-voltage requirements, cooling direction, and service access.
The safest purchasing method is to obtain a validated compatibility list or approve the PSU, PDB, and enclosure together.
A CRPS power distribution board is the intermediate server power assembly that receives high-current power from one or more redundant PSU modules and distributes it to the motherboard, processors, GPUs, storage, fans, and auxiliary devices while potentially handling protection, current sharing, monitoring, standby power, and downstream voltage conversion.
Its connector type and current rating should be treated as part of the chassis specification, not as a generic accessory.
The required CRPS power supply capacity is the maximum validated system demand—including CPUs, GPUs, memory, drives, fans, motherboard loads, expansion cards, conversion losses and transient margin—calculated at the actual facility input voltage and again under the intended failed-PSU condition if the server requires redundancy.
Do not calculate from advertised PSU wattage alone because some supplies deliver substantially less output at lower AC input voltages.
The best server chassis for a CRPS power supply is one validated for the exact PSU and PDB configuration, with sufficient mechanical clearance, current capacity, front-to-rear airflow, hot-swap extraction space, motherboard and GPU power routing, rack depth, management compatibility, and redundancy capacity for the server’s maximum intended workload.
For OEM projects, I would prioritize documented compatibility and accessible engineering drawings over a generic claim that the enclosure “supports redundant power.”
To verify CRPS PSU compatibility, send the chassis manufacturer the exact PSU manufacturer and model, mechanical drawing or STEP file, PDB specification, mating connector information, facility input voltage, system peak load, redundancy requirement, motherboard and GPU configuration, airflow direction, rack depth, management requirements, and expected production configuration.
The more completely those inputs are frozen before CAD begins, the lower the probability of redesign after prototyping.
Do not buy a server chassis because the product page contains the word CRPS.
Verify the system.
Send your chassis supplier:
Then request confirmation of mechanical fit, electrical interface, PDB capacity, redundant operation, airflow, hot-swap access, and full-load behavior before the prototype is approved.
For an OEM or custom project, have the server chassis engineered around the real power subsystem from day one. That is considerably cheaper than discovering during integration that a “CRPS-compatible” chassis and a “CRPS-compatible” power supply were never truly compatible.
Mark Lee is the founder of ISTONECASE, with 20 years of experience in the server chassis industry. He specializes in OEM/ODM solutions for GPU and AI, rackmount, industrial, wallmount, NAS, Mini-ITX and multi-node chassis. His expertise supports customized hardware projects for data centers, AI computing, enterprise storage, edge computing, networking and industrial applications.
Comments
