What does it cost?
No rates are published yet. The model is by the hour: you pay for the hours compute is held for your job, the meter stops when the job finishes, and nothing is owned afterwards.
· 3 min read
IN SHORT
There is no published rate today, and the preview says so wherever cost comes up. The design is by the hour: compute is held for a job from rent to release, the meter runs for those hours, and it stops when the job finishes. Time spent waiting in a queue is not metered. Rates would be published when access opens, not before.
No one can tell you today, because no rate has been published. E31 Network is a technical preview, nothing has shipped, and the price list does not exist yet. What is published is the shape of the bill: how the meter would run, when it would stop, and what would never appear on it.
That shape is worth understanding even without the number, because it is the part that decides whether the network fits a team at all.
How the meter would run
The design intent is to charge by the hour, for the hours compute is held for a job. The meter starts at the rent step, when compute is set aside for your work, and it runs through the run step while the job executes unattended. It stops at release, when the output has been returned and the compute goes back to the pool.
Those two steps, rent and run, are the whole of what you would pay for. The submit step costs nothing. The output step is the return of your results. The release step is the meter stopping. The life of a job is drawn out on how it works, and the paid portion is a clear band in the middle of it.
The illustrative requests on the examples page are written in this currency. "I need eight GPUs for four hours to process this dataset" describes how much compute and for how long. Multiply that by a rate and you have a bill. The rate is the missing factor.
What you would not pay for
Queue time. A job that can wait goes into a queue until compute is free, and the meter does not run while it waits. This is the design decision that makes the network different from owning a machine: the waiting is free because nothing is being held for you yet.
Idle time afterwards. When the job finishes, the compute is released. There is no instance left running overnight, no reservation to remember to cancel, no hardware depreciating in a closet. Nothing is owned, so nothing sits idle on your bill.
That is the economic argument in one line. A team whose need is heavy some weeks and light most weeks pays for the heavy weeks only. Owning enough compute for the peak means paying for it at the trough. The question of whether you need any hardware at all has its own answer at do I need to own any hardware.
Why there is no number yet
Where the capacity comes from is not published, and a rate depends on it. Scheduling mixed job sizes fairly is on the program's list of unsolved problems, and a rate that ignores that problem would be a guess. The program has chosen to publish the mechanism now and the price later, rather than publish a price it would have to walk back.
When access opens, rates would be published with it. Until then, any figure you hear attached to E31 Network did not come from E31 Network.
Questions this raises
- Will there be a minimum charge or a minimum job size?
- Nothing of the kind is published. The design describes a meter that runs while compute is held and stops at release, and no minimum, commitment, or tier has been added to that description. If one were needed, it would be published alongside the rates.
- Would a job that fails partway through still be charged?
- The design meters the hours compute was held, and a job that fails still held compute until it stopped. How failures are accounted for in detail is not yet published, and it is a fair question to bring to the program.
- Can I estimate my cost from the illustrative requests?
- Not in dollars, because there is no rate to multiply by. You can estimate the hours. A request for eight GPUs for four hours describes its own meter; the rate it would be multiplied by is the part that does not exist yet.
KEEP READING