Who it is for
A page for each reader.
ADAPT is one per-GPU software runtime, and each reader meets it from a different side of the power envelope. Find your situation below, then read what you are measured on and what released capacity is worth on your numbers.
"Our book is sold ahead of our megawatts. The racks are on the dock and the power is not."
For GPU cloud operators: a per-GPU runtime measured at up to 21% less GPU die power on H100 NVL, so contracted capacity fits the megawatts already held.
"Our halls were designed for a lower density than the kit arriving in them."
For colocation: the tenant runs ADAPT on its own accelerators, measured at up to 21% less GPU die power on H100 NVL, and the operator keeps halls that fit.
"We run our own accelerators, and the building has become the limit."
For enterprise estates: more of the work in the queue through the same feed, from a runtime measured at up to 21% less GPU die power on H100 NVL.
"The customer wants the system. The site cannot power what we quoted."
For systems builders: more of a quoted configuration fits the customer's power, modelled from a runtime measured at up to 21% less GPU die power on H100 NVL.
"The envelope was fixed when the machine was funded, and the queue has outgrown it."
For research computing: more accelerator-hours inside a funded power envelope, from a runtime measured at up to 21% less GPU die power on H100 NVL.
"The compute is funded, the envelope is fixed, and the connection date is not ours to move."
For public bodies: more research capacity inside a funded power envelope, from a runtime measured at up to 21% less GPU die power on H100 NVL.
Also reading
Your own numbers
Start from your own envelope.
The instrument reads released capacity on your hardware and your power. The case studies work situations through to racks and megawatts. Or tell us what you run.