One Frame, Every Workload
The universal rack, and where the copper goes
Why four very different machines share one welded frame. What converting at the top of the rack saves. Why the heavy conductor should be short.
Abstract
A data centre floor holds a mainframe drawing under seven kilowatts, a colocation rack at fifteen, and liquid-cooled accelerated compute at a hundred and fifty, often in the same row. This paper sets out why HyperNext built one welded frame to hold all of them rather than buying three rack families, and why the decision that matters most electrically is not the voltage but where along the path the conversion happens.
Contents
- 011. The problem with buying three rack families
- 022. One frame, four models
- 033. Where the copper goes
- 044. Converting at the top of the rack
- 055. Air today, liquid tomorrow
- 066. Built to Indian standards, and what that meant
- 077. The parts nobody specifies
- 088. References and standards
Request the full paper
The complete paper, including all figures, tables, references, and citations, is available as PDF. Enter your details to receive it.
Request paper · HN-RP-016.pdfKey findings
- Conductor area for a given power and loss goes as the inverse square of the voltage. Distributing one 150 kW rack at 50 volts rather than 800 needs about 256 times the copper over the same distance.
- Converting at the top of the rack keeps the 3,200 ampere conductor 2.2 metres long instead of thirty. The high-current run is short by design, not by accident.
- One welded frame carrying 2,200 kg static and 1,500 kg rolling serves four models. What changes between them is the kits bolted to it, not the structure.
- Racks convert between models without cutting, welding or re-coating, so a hall can be repurposed from storage to accelerated compute without replacing its enclosures.
- Following EIA-310 and Open Rack V3 in the same frame costs mounting complexity and buys the ability to put any tenant in any position on the floor.
