Whole GPUs by the hour. Clusters when you need them.
TrainPods are on-demand and reserved NVIDIA GPUs, whole cards and clusters, billed by the hour, for training runs that need the full machine.
TrainPods are on-demand and reserved NVIDIA GPUs billed by the hour. Boot an instance, connect over SSH with the podstack CLI, and train on the full card or a multi-GPU node.
TrainPods at a glance
What TrainPods does
Instant-boot instances
On-demand instances with A100, H100, H200, B200 and L40S GPUs boot in seconds and are ready for SSH.
Reserved clusters
Dedicated multi-node clusters reserved for weeks or months, for training runs that need predictable capacity. Offered where partner capacity exists.
Bring your own stack
A TrainPod is a full machine. Install your framework, mount your data and run your own scripts.
CLI-first workflow
The podstack CLI provisions instances, opens SSH sessions and syncs files, so training runs script cleanly.
From the terminal
One CLI, one SDK, one API
$ podstack gpu launch
? pick GPU › H100 80GB · 50% slice
✓ ssh ready · billed per GPU-minute
$ podstack train create --budget 25
→ LoRA job started · adapter saved when it finishes
$ podstack models list
llama-3-70b · mistral-7b · sdxl · +18How TrainPods works
- 1
Pick a GPU and region
Choose the GPU family, the number of GPUs and the region closest to your data.
- 2
Boot
On-demand instances start immediately. Reserved clusters are confirmed with a start date.
- 3
Connect and train
SSH in with the podstack CLI, bring your code and data, and run.
- 4
Release
Stop the instance and hourly billing stops. Reserved capacity runs for its agreed term.
How TrainPods bills
per hour
On-demand instances bill hourly from your Podstack wallet. Reserved clusters are quoted for their term; request one and a quote comes back.
See pricingQuestions about TrainPods
What is a TrainPod?
A TrainPod is a whole NVIDIA GPU instance or a reserved multi-GPU cluster, billed by the hour, that you connect to over SSH.
Which GPUs can I rent?
TrainPods offer A100, H100, H200, B200 and L40S GPUs, with availability shown live in the Training Cloud console.
How do I connect to a TrainPod?
You connect over SSH using the podstack CLI, which also syncs files and manages the instance lifecycle.
Can I reserve a cluster for a long run?
Yes, reserved clusters are available for weeks or months where partner capacity exists, and a request returns a quote.
How is a TrainPod different from a QuickPod?
A QuickPod is a fractional-GPU stack launched from a template, while a TrainPod is a whole GPU or cluster you manage as a full machine.
When am I charged?
On-demand TrainPods bill per hour while running and stop billing when you stop them; reserved clusters bill for their agreed term.
Start with TrainPods
TrainPods are on-demand and reserved NVIDIA GPUs, whole cards and clusters, billed by the hour, for training runs that need the full machine.