Buyers ask two questions about rack load, and they are different questions. The first is whether the frame holds the cluster being ordered this quarter. The second is whether it still holds that cluster in year eight, on a slab it does not control, in a building that may have to survive a seismic event. A catalogue rating answers neither on its own — it is a claim until a test log, a material gauge and an anchorage detail sit behind it.
What 1,600 kg actually rates
The rating is a uniformly distributed static load carried by the frame in its installed position: rails, uprights, base plinth and levelling feet, with the load path closed through the floor. On the RackForge line the number splits by family:
- Standard 42U–52U cabinets: 1,500 kg static and 750 kg dynamic in service, in 600 mm and 800 mm widths.
- GPU high-density frames: 1,600 kg static across the same 42U–52U heights, 800 × 1200 mm as standard, with 600 mm builds quoted for NVL72 reference alignment.
- Dynamic transport build, optional: 950 kg — for rows that arrive loaded and are moved across a hall, or shipped racked.
The distinction between static and dynamic is not marketing language. A frame that never leaves its bay sees a stationary load; a frame that is wheeled across a floor, lifted onto a dock or rolled into a container sees that load multiplied by every jolt in the journey. Quoting one number for both conditions is how overruns happen, which is why the two ratings are published separately.
Why the safety factor is 4×, not 1.5×
A safety factor is a ratio between what the structure was proven to carry and what it is rated to carry in service. Every kilogram of that margin buys a specific protection:
- Load estimate error — buyers rarely know final rack weight at order time. A 48U frame filled with liquid-cooled trays, manifolds, cabling and power shelves is estimated before the last three of those are chosen.
- Concentration — the distributed rating is not a point-load allowance. A single 4U chassis of 60 kg on two rails loads the uprights differently than the same mass spread over 20U.
- Installation abuse — rails are stood on, frames are levered, bayning brackets are tightened by feel. A 4× margin absorbs a mistake that a 1.5× margin turns into a bent column.
- Time — the reason for a 72-hour hold is that steel keeps moving after the load is applied. The proof test is not finished when the weights land.
Concretely, the outbound record for a high-density frame reads: a 4× static hold at 6,400 kg for 72 hours, then a creep recheck with deflection measured against a 1.0 mm limit, then an ISO 2409 adhesion cross-cut on the coating and a ground-continuity measurement. The outbound QC log ships inside the crate, serial-matched to the frame.
Creep: the part of the rating that takes days
Elastic deflection appears instantly and reverses when the load leaves. Creep is the slow, permanent component: under sustained load, the frame keeps settling as material yields locally at bolt holes, weld toes and bend radii. On a rack that is loaded once and left for a decade, creep — not yield — is the mechanism that decides whether the top U of the row is still square in year eight, and whether a door still shuts on a warped bay.
Reading a creep record is a three-point exercise. Note the deflection at the moment full load lands, note it again at the end of the hold, and compare the drift against the acceptance limit. On the RackForge high-density frames, the recorded figure is 0.8 mm of deflection inside a 1.0 mm limit after the 72-hour hold. Anything that keeps drifting at hour 72 has not stabilized, and an unstable frame is a frame whose rating is provisional.
Two design choices do most of the work in holding that number down. Uprights are five-bend formed in a single setup on 2.0 mm SPCC, so stiffness comes from geometry rather than extra gauge and the bend lines stay parallel to ±0.1 mm down the whole column. Bayning brackets — parallel and closure — are fitted at assembly, which turns a row of neighbouring frames into one laterally braced structure instead of a line of independent posts.
Seismic is not a heavier static load
The most expensive misconception in rack procurement is treating a seismic requirement as a static rating with a bigger number next to it. The load path is different, the failure modes are different, and the parts that must be tested are different:
| Demand | What it loads | Where the capacity comes from | How it is proven |
|---|---|---|---|
| Static load | Uprights in compression, rails in bending, feet in bearing on the slab | Five-bend 2.0 mm SPCC columns, rail gauge, base plinth | Weighted-plate proof at 4×, 72 h hold, creep recheck |
| Transport shock | Rails and bayning joints under vertical and lateral jolts | Optional heavy build rated 950 kg dynamic | Build-level verification for loaded transport |
| Seismic motion | Anchors in shear, base in tension and overturning, row in lateral drift | Bolt-down kit, base stiffening, bracing | Zone 4 kit engineering with slab and anchor calculation |
Frame-to-frame ground continuity is measured at ≤0.1 Ω on the same build, with bonding jumpers factory-fitted and logged, because a rack row is also part of the electrical reference for everything installed in it.
GR-63-CORE Zone 4 as an orderable option
Zone 4 is the most severe seismic zone in the GR-63-CORE environmental envelope, and RackForge offers it as a bolt-down kit ordered with the rack and factory-drilled rather than as a blanket rating printed on every frame. That wording matters: a standard frame that has not been ordered with the kit is not a Zone 4 frame, and a Zone 4 kit fitted to a floor that cannot take the anchor loads is not a Zone 4 installation either. Three checks decide the outcome:
- Anchorage to the slab. The kit's holding capacity is only as good as the concrete under it. Slab thickness, reinforcement and anchor embedment are site inputs, and the anchor calculation belongs in the project's structural package, not on the rack drawing.
- Overturning. A tall 52U frame with a heavy top-of-row tray mass has a centre of gravity well above its base. Seismic demand resolves into a moment that the anchors must resist in tension on the leading side of the frame.
- Row bracing. Bayning brackets tie neighbours together; on NVL72-class builds the optional NVLink braced frame adds a diagonal set that stiffens the bay against large tray masses. Both are fitted at assembly, not improvised on site.
Order the kit at the same time as the frame. The drilling pattern is set at the punching stage, and a bay that arrives without it is a bay that gets field-drilled — through a coated surface, above a live aisle, by whoever is available.
The three numbers to demand before you order
- The rating and its test basis. Not just "1,600 kg", but the proof load (6,400 kg), the hold duration (72 h) and the acceptance limit on creep (1.0 mm). Ask for the log format before the purchase order, not after the frames land.
- The material gauge on the load path. Columns are 2.0 mm SPCC with five-bend forming; doors and panels are 1.2–1.5 mm. A rating quoted without a gauge and a geometry note is a rating that cannot be checked.
- The floor interface. Static load per cabinet against the slab's rated capacity, plus the transport route — door clearances, lift capacity and any ramp — checked before the first frame is uncrated.
The full structure, the standard and high-density load records and the Zone 4 option are specified on the GPU high-density rack page, and the ratings for the general-purpose family sit on the standard server rack page. Both are built on the same line, which is why the load figures in this note and the load figures on those pages are the same figures.
