A 2U chassis is three problems that have to be solved at the same time. It has to fit a card cage, a power supply and a backplane into 88.9 mm of rack height. It has to move the heat back out of a box that was made as full as it could be. And it has to let a technician pull one module out while the rest of the rack stays in service.
We see the cost of solving those three separately. A chassis passes its bench thermal test with the lid off, then measures 18 K hotter in the rack, because the unit below is exhausting straight into its intake. A design adds a divider to stiffen the card cage and quietly removes the cross-section the air was using. A module is sized to the millimetre of available width, and the first field replacement becomes a two-person job with the unit unracked.
None of those are fabrication faults, and none of them are caught by a drawing check. They are decisions taken in the first week of layout, and this guide sets out the three pieces of arithmetic and the two interfaces we ask for before we cut anything — because a chassis that is easy to build and hard to keep running is a worse product than one that is a little harder to build.
Key takeaways
- 3.5 inches is a pitch, not a budget. The nominal height belongs to the rack. What you actually spend is depth, card pitch and the clearance around them.
- Depth is the dimension no standard fixes. It is negotiated with the cabinet, the rear cable bend and the door — not with the chassis drawing.
- Cooling is arithmetic, not a fan size.
V̇ = Q ÷ (ρ · cp · ΔT), which is roughly 70 CFM per 600 W at a 15 K rise. - A punched panel is an orifice. A panel that is 45 per cent open geometrically behaves like about a quarter of its face area in flow terms.
- Fan curves are measured on a free inlet. Every louvre, filter and perforated band you add moves the operating point down the curve.
- Racked service is designed in or paid for forever. Clearance for a module to slide, and access to the test points, are cheap at layout and expensive later.
The space budget: what 3.5 inches actually buys
Two rack units are 3.5 inches, or 88.9 mm, of vertical pitch. Inside that you have to place a card cage, the power supply, the backplane, the wiring and the air. The order in which you spend it matters more than the total, because the first decision constrains the other three.
Card pitch: the decision everything else inherits
Card pitch is not a function of board thickness. It is the largest of four separate requirements: the connector's keep-out, the clearance your electrical design needs between conductors at the working voltage, the cross-section you have decided the air gets, and the room a technician needs to get a connector off without levering against a neighbouring board.
The mistake we see most often is deriving the pitch from the PCB and then looking for the structure afterwards. Fix the air gap first — it is the hardest of the four to recover later, because recovering it means respacing the whole cage.
Depth: the dimension no standard fixes
The rack standards fixture the width, the hole pattern and the unit height. They do not fix depth, which is why two 2U chassis can be 300 mm and 700 mm deep and both be compliant. Depth is negotiated with three other things: the rear cable bend, the cabinet door or rear ducting, and whether the chassis is loaded from the front or slid out on rails.
Give the rear harness a stated minimum bend radius rather than whatever is left. Ask the cable supplier for the figure: it is normally given as a multiple of the cable's outer diameter, and the multiple is different for a cable that is installed once and a cable that flexes in service. Where a service loop is unavoidable, define a tie-off point so the excess sits out of the intake instead of in front of it. The dimensions the rack standard does fix are set out separately, including the hole patterns and the tolerance stack-up over a tall assembly.
Front panel: cutouts, indication and pull relief
The front panel carries the connectors, the indicators, the handles and the latches, and it is the face that has to survive being built, racked, cabled and serviced. Size connector cutouts from the connector manufacturer's panel drawing rather than from the connector's own body, and carry the assembly tolerances through to the finished part.
Two details decide whether the panel is usable in a populated rack. The first is whether the indicators and handles are still readable and reachable when the unit has equipment directly above and below it — a design that looks clean on a bench can be unreadable in the middle of a 42U frame. The second is pull relief: the panel and its fasteners must survive a cable being pulled hard, without the load transferring into the PCB behind it.
Where the rear panel can carry the secondary I/O and the power inlet, do that. It shortens the internal runs, keeps the front face clean and removes bending from the harness that runs past the cards.
The cooling path: from heat load to hole pattern
Almost every 2U chassis in this class runs front to back: cool air enters at the front, crosses the cards, and leaves hot at the rear. The intake and the exhaust therefore have to be aligned to that path as a pair. Punching the same vent pattern into both panels and hoping the air takes the hint is the origin of a large share of the thermal problems we are asked to look at.
Start from the heat load, not from the fan
The airflow the chassis needs is set by how much heat has to leave and how much temperature rise the components will accept. For air, that relationship is short:
V̇ = Q ÷ ( ρ · c_p · ΔT )
V̇ volumetric flow (m³/s)
Q heat to be removed (W)
ρ air density, ~1.2 (kg/m³)
c_p specific heat, ~1005 (J/kg·K)
ΔT allowable rise above the intake air (K)Air density times specific heat is about 1,206 J per cubic metre per kelvin, which is a convenient number to carry around. At a 15 K rise the chassis needs roughly one cubic metre per minute for every 300 W:
| Heat to remove | Flow at ΔT = 10 K | Flow at ΔT = 15 K | Flow at ΔT = 20 K |
|---|---|---|---|
| 150 W | 0.0124 m³/s · 26 CFM | 0.0083 m³/s · 18 CFM | 0.0062 m³/s · 13 CFM |
| 300 W | 0.0249 m³/s · 53 CFM | 0.0166 m³/s · 35 CFM | 0.0124 m³/s · 26 CFM |
| 600 W | 0.0497 m³/s · 106 CFM | 0.0332 m³/s · 70 CFM | 0.0249 m³/s · 53 CFM |
| 1,000 W | 0.0828 m³/s · 176 CFM | 0.0552 m³/s · 117 CFM | 0.0414 m³/s · 88 CFM |
Two cautions before the number is used as a fan specification. It is the flow the chassis has to pass, not the flow the fan advertises, and it assumes the whole rise is available — if the intake air is already 10 K above ambient because the unit below is exhausting into it, the arithmetic has just told you to fix the rack, not to buy a bigger fan. That is one of the reasons vent pattern design and open-area targets are worth settling before the fan is chosen.
Turning flow into a hole pattern: two velocities
Once you know the flow, the panel becomes a fluid problem with two numbers in it.
Face velocity is the flow divided by the free area of the vent. It is how fast the air has to move through the holes, and it is the number that decides whether the panel is a vent or a restriction.
Velocity pressure is what that speed costs you: ΔP = ½ ρ v². The relationship is quadratic, so doubling the velocity quadruples the cost. It is also why velocity is the term to control: at 3 m/s the velocity pressure is about 5 Pa, at 6 m/s about 22 Pa, and at 10 m/s about 60 Pa — and that last figure is a large fraction of what a low-profile 2U fan can still deliver.
Then correct the area, because a punched hole is an orifice rather than a tube. A sharp-edged hole passes less air than its geometric area suggests; a discharge coefficient near 0.6 is the usual starting point. Put the two effects together and a panel that is 45 per cent open geometrically behaves like roughly a quarter of its face area in flow terms. Perforated plate buys a lot of stiffness and cosmetic uniformity, and it costs you this factor, which is why open area is a design target and not a cosmetic choice.
Two further effects move the answer in the same direction. A louvre or a filter element adds an entry loss on top of the orifice loss. And the fan itself is characterised on a free inlet: the panel, the filter and the cards all add system resistance, which moves the operating point down the fan curve, so a fan specified at 70 CFM in free air does not pass 70 CFM through a real 2U front panel. Ask the fan supplier for the curve rather than the headline figure, and plot the panel's resistance against it.
Baffles and dividers: spending air where it is needed
A card cage with a clear straight path through it will move most of its air through the easiest gap, not through the hottest card. Internal baffles and dividers exist to break that shortcut: they force the stream across the components that need it rather than past them, and they stop the recirculation where hot exhaust turns back along the inside of the top cover.
Two consequences for the panel design follow. First, a baffle is a pressure drop, so it belongs in the resistance calculation from the start — not discovered after the first thermal test. Second, the practical limit is usually the return edge of the baffle: keep it off the fan's impeller and out of the module slide path, or the cooling feature turns into an assembly and service problem.
Passive cooling, and where it stops working
Below a certain dissipation, natural convection is enough: a chassis with a continuous vertical path from a low intake to a high exhaust, no horizontal shelf blocking the rise, and enough vent area that the face velocity stays low. That route suits 2U units that are almost entirely passive electronics, and it removes the fan and its filter from the maintenance schedule.
What passive cooling cannot do is be added later. A design that needs a fan to pass a thermal test, and is then shipped with a louvre band instead, has not been cooled; it has been delayed. Decide the cooling mode at layout, because it sets the vent areas, the internal clearances and the position of every heat-generating part.
Serviceability: design for the callout, not for the install
In service, a rack unit is touched far more often by someone maintaining it than by someone installing it. That asymmetry has a direct design consequence: the decisions that look like generosity at layout — clearance, accessibility, a standard mounting interface — are the ones that decide the real cost of ownership.
Racked service and tool-less access
Ask what has to be done with the unit in place: swap a drive, reach a fuse, reseat a connector, read a status indicator, attach a test lead. For each of those, decide whether it is reachable from the front, from the rear, or only from the top. Top access is the expensive answer, because it needs either a large clearance below the unit above or a slide-out design, and the slide-out brings its own rail and clearance constraints.
Thumb screws, captive fasteners and tool-less latches are worth more than they cost. So is a consistent fastener type across the unit: a technician who has opened one of your chassis should not need a second tool for the next one.
Hot-swap modules and slide clearance
Anything described as hot-swappable — a fan tray, a power supply, a drive carrier — has to come out without disturbing its neighbours and without pulling on a cable that terminates elsewhere. That means the slide path is a real design element, with a defined clearance to the adjacent module and to the baffles, and a connector that releases without the module having to be tilted into the space it does not have.
The connector choice usually decides this. If the module has to be levered, the clearance you designed is being used as a tool, and the mating cycle count becomes a service-life figure rather than a specification.
Fastener discipline and captive hardware
Keep the fastener count low and the hardware captive wherever a screw could fall into the chassis or behind the rack. A dropped screw in a populated frame costs far more in technician time than the hardware ever saved, and in a live unit it can become a conductive foreign object rather than an inconvenience.
The same discipline applies to the parts that come off during service: define which panels are removable, keep the fasteners on the panel, and do not rely on the cable harness to remember where a cover goes.
Test points and the upgrade path
Leave the measurement points reachable with the unit racked, and label them. A technician who can identify a fault in seconds and swap the right module differs from one who has to unrack the unit to confirm a hypothesis.
The last serviceability decision is the cheapest one to make and the least often made: agree a standard mounting interface inside the chassis. If cards, drives and power supplies all mount on a repeated pattern, a later revision replaces the insert rather than the chassis. That is the difference between a version change and a re-qualification.
Where the three decisions collide
Every item in the table below is a genuine gain in one column and a real cost in another. The point of listing them together is that they are usually decided in three different meetings.
| Decision | Buys you | Costs you | Settle it by |
|---|---|---|---|
| Stiffening divider in the card cage | Rigidity, and a shorter unsupported span | Airflow cross-section, and pressure drop | Putting the divider in the resistance calculation, not adding it after a test |
| Close card pitch | More slots in the same 2U, or a shorter chassis | Air gap, connector access, and electrical clearance | Stating the air gap first and letting the slot count follow |
| Perforated vent band instead of a louvre | Cosmetic uniformity and panel stiffness | Effective flow area, and depth of the opening | Comparing effective area at the required face velocity |
| Filter on the intake | Dust control, and a longer service interval | Fan operating point, and a maintenance item | Deciding who replaces it and how often, before it is specified |
| Slide-out trays and rails | Full access without unracking | Depth, and clearance to neighbours | Confirming the cabinet depth before the chassis is released |
| Tool-less latches and captive hardware | Faster, safer service | Part count, and a small per-unit cost | Counting what a field callout costs against the hardware |
| Standard internal mounting pattern | A cheaper future revision | Some layout freedom now | Deciding whether a second revision is likely before the first is frozen |
How we build a 2U chassis
We fabricate 2U chassis and rack-mounted sheet metal parts from customer drawings. The manufacturing review happens on the drawing, before any material is cut, because that is the only point at which a finding is still free.
What the review looks for
The engineering team reads the model for the things that decide whether the first article is right: card pitch against the connector keep-outs, panel depth against the rear bend radius, bend radii that match the tooling, weld access on the seams that will be welded, and fastener placement against the underlying structure. Where the drawing leaves a vent or a cutout positioned only by eye, that comes back as a question rather than as an assumption.
Materials
Cold-rolled steel such as SPCC or Q235 is the default for indoor equipment. Stainless steel 304 or 316 is specified where corrosion drives the decision, including coastal and wash-down locations. Aluminium is used where the chassis itself is part of the thermal path, or where mass matters, and galvanised steel sits between cold-rolled and stainless as a corrosion compromise. Thickness follows the load path and the deflection limit you set on the rack ears — we ask for that limit rather than assuming one. The alloy comparison and its service-environment rules are set out separately.
Process chain and first-article verification
A chassis of this kind is cut, punched, formed, welded and finished on our own line — nine in-house operations covering drawing review and DFM, laser cutting, precision bending, forming, welding, grinding and deburring, riveting, electrostatic powder coating, and assembly with QC. Critical dimensions on the first article are checked on a CMM before the part is released, and the 5,000 m² facility runs under ISO 9001 and ISO 14001 quality and environmental management, with RoHS and REACH compliance on the materials we supply.
What we do not do is test your thermal design. We build the vent geometry you specify and we will tell you when the arithmetic does not look plausible, but the airflow measurement, the compliance position and the ingress rating remain yours or your test house's. We do not issue compliance declarations.
What to put in a 2U chassis RFQ
Most 2U quotes come back accurate on the fabrication and qualified on everything else, because the drawing does not contain the four numbers that decide the design. Send these with the model:
2U CHASSIS RFQ — [PROJECT / PART NUMBER]
1. Heat to be removed (W) and the allowable air temperature rise (K)
2. Available depth inside the cabinet, and the rear cable bend radius
3. Rack location: rail type, what sits directly above and below, and whether
the unit is front-loaded or slide-out
4. Service access: what is replaced in the field, and from which face
5. Cosmetic class and finish, in the terms your own specification uses
6. Expected second revision: is a change to the internal layout likelyItem 1 turns the vent pattern from a drawing detail into a calculation. Item 3 is what tells us whether the intake will be drawing pre-heated air. Item 6 is what tells us whether to make the internal mounting pattern standard now or to optimise the layout for this version only. The commercial fields around that block — quantities, delivery, documentation — are the subject of what belongs in an enclosure RFQ.
> Send the model and those six items, and we will return a manufacturing review — including the questions we would ask about the vent area — before quoting. Minimum order quantity is 1 piece with no tooling charge on the sheet metal route, standard parts ship in 7 to 15 working days and custom parts in 20 to 30 days. Get the review started.
Frequently asked questions
What does 2U mean on a rack chassis?
One rack unit is 1.75 inches, or 44.45 mm, of vertical space in a 19 inch rack, so 2U is 3.5 inches (88.9 mm) nominal. That number describes the pitch the rack reserves. It says nothing about the depth, the card layout or the cooling path, all of which you design.
How do I size the vents on a 2U chassis?
Start from the heat load, not from a fan. The airflow the chassis needs is the dissipation divided by the product of air density, specific heat and the temperature rise you are willing to allow. Work the result into a face velocity, then compare the velocity pressure it implies against the static pressure the fan can still deliver at that flow. A panel that is 45 per cent open geometrically does not behave like 45 per cent of the face area, because a punched hole is an orifice.
Can a 2U chassis be cooled without fans?
Yes, when the dissipation is low enough that natural convection can carry it with a temperature rise the components tolerate. The requirement is a continuous vertical path from the intake to the exhaust with no shelf of cards blocking it, and enough vent area that the velocity pressure stays small. Above that, a fan is not optional, but its position, cutout and inlet clearance become part of the design rather than an afterthought.
What materials are used for a 2U rack chassis?
Cold-rolled steel such as SPCC or Q235 is the default for indoor equipment, because it is stiff per unit of cost and takes a powder coat well. Stainless steel 304 or 316 is chosen where corrosion drives the decision, including coastal and washdown locations. Aluminium is used where the chassis itself is part of the heat path or where mass matters. Galvanised steel sits between cold-rolled and stainless as a corrosion compromise. Thickness is set by the load path and by the deflection limit you place on the rack ears, not by a standard.
Can a 2U chassis be customised, and what is the minimum order?
Yes. We fabricate 2U chassis and rack-mounted sheet metal parts to customer drawings, with a minimum order quantity of one piece and no tooling charge on the sheet metal route. Send the model together with the heat load, the available depth and the service access you need, and we return a manufacturing review before quoting.
What lead time should I expect for a custom 2U chassis?
Standard parts ship in 7 to 15 working days and custom parts in 20 to 30 days, with sample lead time quoted by complexity rather than as a single figure. The lead time that actually moves a project is the one before the order, so the review above is deliberately done on the drawing rather than on the first article.