01 / Match
We pick
your GPU.
We compare seven providers and choose the machine with the lowest cost to finish your job.
What countsHourly price, GPU speed and how often the machine gets preempted.
GPU requirements
GPU jobs across 7 providers.Match / Cost / Complete
For training, fine-tuning and batch inference
Train AI models
on the cheapest GPUs.
Give it your training job and a budget.
How it works
You bring the job.
We handle the GPUs.
Send your container, GPU memory and budget through one API. We pick the machine and run the job. The same API shows status, spend and saved checkpoints.
We pick
02 / Cost
We stay
inside your budget.
Spend is tracked while the job runs. At your limit the job stops and its checkpoint is kept.
Before you submitGet a quote for every matching machine, with estimated hours and cost.
Budget reference
03 / Complete
We restart
interrupted jobs.
If a machine is taken back, the job moves to the next one and resumes from its last checkpoint.
What your code needsSave checkpoints to /outputs and load the newest one on start.
Checkpoint guide
Python SDK
Submit and track jobs
with the Python SDK.
Run runcompute login, then submit a container with your code. The job you get back shows status, spend and saved files.