CLUSTER
4× 8-GPU nodes
UNIVERSITY
Universities — department-scale compute
GRANT = ONE LINE ITEMMIXED VENDORS OKTEACHABLE INSTRUMENT
A lab or department can own serious GPU compute without the university building them a machine room.
TDP × 1.2 · room = 1.1 × IT
IT LOAD (TDP × 1.2)
30.4 kW
ROOM MUST EXTRACT
33.4 kW ≈ 9.5 tons
DELIVERY
1 cabinet
A full department cluster in one cabinet — squarely inside the 20–50 kW gap. The room-load worksheet is part of the grant paperwork; we supply it.
Grant-friendly
One line item: cabinet + cooling + monitoring. Bring your own liquid-ready servers as budgets allow — this year H100s, in three years whatever wins.
Open hardware
ORv3 blind-mate + UQD: mixed generations and vendors in one cabinet. The university is never locked to one node vendor — check any competitor on this point.
Teachable
Instrumented loops and published thermal data make it a lab instrument, not a black box — students read real derating curves, not marketing charts.
We do not promise compatibility with a specific server model until its UQD/DLC interface is confirmed against the manifold spec — send the model list, we confirm each one in writing.