Skip to content
Use cases
Computeruns today
Compute

Train a model without buying hardware

Take an H100 for the length of a training run and give it back. No reservation, no minimum, and no card sitting idle between experiments.

A training run needs a large card for a few days and nothing for the weeks either side. Buying for the peak means paying for the trough.

Why it is hard

  • The card is idle most of the month

    Training is bursty. A GPU bought for the heaviest week of the quarter spends the rest of it depreciating, and the finance case has to carry all of that.

  • Reservations outlive the experiment

    Most providers want a commitment measured in months. Experiments are measured in days, and the ones that fail are measured in hours.

  • You cannot try a bigger card cheaply

    Finding out whether a run needs 80GB or fits in 48 usually means committing to the answer before you have it.

What answers it

  • Take it for the run, give it back

    Metering starts when the container does and stops when it stops. There is no minimum and no notice period, so an aborted run costs what it used.

  • No reservation at all

    Capacity is whatever providers have online right now. A machine already rented is not offered to you, so what you see is what you can have.

  • Change your mind for the price of an hour

    Seven GPU models, from a 24GB 4090 to a 141GB H200. Trying the larger one costs the hour it takes to find out.

How billing works

usage_eventsbilled per second, EURlaunchmeter starts with the containerheartbeat10s at the price frozen at matchstopmeter stops, nothing after
heartbeatsent twiceusage_eventsledger_entriescharge · provider_share 85%account_balancesa view: sums, never a columnone fresh row, summedthe retry: unchangedunique(instance_id, seq)

per second, to the microeuro · no minimum · no egress fee · idle costs nothing

Metering starts when the container does and stops when it stops. The rate is frozen the moment you match, so a provider re-pricing cannot change a running rental, and an aborted run costs what it used.

Pick a card, paste an SSH key, and point a training script at it.