Table of contents

Pricing & Billing

TrainPod GPUs bill per hour for as long as an instance runs, drawn from your Podstack wallet balance. Managed fine-tuning jobs bill per run, which you can cap with a budget. This page explains how to check prices, fund your wallet, and control spend.

Check live prices

Prices vary by GPU type, region, and tier. Query them with:

podstack gpu pricing
podstack gpu pricing --gpu-type h100_sxm
podstack gpu pricing --gpu-type h100_sxm --tier on_demand

GPU TYPE    COUNT   REGION        TIER        PRICE          AVAIL
h100_sxm    1       asia-pacific  on_demand   ₹405.74/hr     4
h100_pcie   1       ca-central    on_demand   ₹203.55/hr     84
a100_80gb   1       ca-central    on_demand   ₹188.62/hr     4
a100_40gb   1       india-south   on_demand   ₹120.77/hr     3
  • PRICE is the final Podstack price per hour (in INR) — all fees are already included. There’s no separate management or egress markup on top.
  • AVAIL is how many matching GPUs can be launched right now, and REGION is where they run.

Prices shown above are illustrative snapshots — always run podstack gpu pricing for the current rates, regions, and availability.

Spot vs on-demand

TierPriceBehavior
on_demandHigherRuns until you terminate it. Best for jobs you can’t afford to be interrupted.
spotLowerCheaper, but can be reclaimed when capacity is needed. Best for interruptible or checkpointed work.

Compare both for a GPU type before launching:

podstack gpu pricing --gpu-type h100

Cap the price with --max-price

When launching on spot (where prices can move), set a ceiling. The instance won’t launch above your cap:

podstack gpu instances create --type h100_sxm --tier spot --max-price 2.00 \
  --ssh-key-id sshkey_abc123

Fund your wallet

GPU time and fine-tuning jobs draw from your wallet balance. Add funds in the Podstack dashboard, and manage payment and balance from the Wallet docs. From the CLI, see Wallet & pricing.

Keep enough balance to cover the full runtime of your instances — if the balance runs low, Podstack warns you and can stop instances to prevent a negative balance.

Low-balance warnings

The CLI surfaces backend billing warnings when you list or inspect instances, so you see them even when piping output. For example:

podstack gpu instances list

⚠  my-trainer: under 5 min of runtime left — recharge or this instance will stop

If you see a warning, top up your wallet before the instance is stopped for non-payment.

Stop paying: terminate instances

Billing continues for as long as an instance is running. To stop charges, terminate it:

podstack gpu instances delete gpu-abc123

Terminating wipes the instance’s local disk. Move results off the box first.

Verify nothing is still running (and racking up cost):

podstack gpu instances list --status running

Fine-tuning budgets

Managed fine-tuning jobs bill per run rather than per instance-hour. Cap the spend on a job with --budget:

podstack train create --model podstack/gemma-4-31b-it \
  --training-file file_123 --budget 25.00

Cost-control checklist

  • Right-size the GPU. Match VRAM to your model with podstack gpu types list; don’t rent an H100 for a job an L40S handles.
  • Use spot for interruptible work, with --max-price as a ceiling.
  • Terminate promptly. podstack gpu instances delete <id> the moment you’re done.
  • Watch for low-balance warnings in podstack gpu instances list.
  • Cap fine-tuning jobs with --budget.
  • Audit regularly: podstack gpu instances list --status running.

FAQs

How am I charged for a GPU instance? Per hour, for the entire time the instance is running, from your wallet. Terminate to stop charges.

Is the price shown the final price? Yes. PRICE in podstack gpu pricing is the final Podstack rate with all fees included.

What happens if my wallet runs out mid-run? Podstack warns you (visible in podstack gpu instances list) and can stop the instance to prevent a negative balance. Top up before that happens; checkpoint spot/interruptible work.

Does spot cost less than on_demand? Usually yes, but spot instances can be reclaimed. Compare both with podstack gpu pricing --gpu-type <type>.

How do I avoid overpaying on spot? Pass --max-price <usd> on podstack gpu instances create so it won’t launch above your ceiling.

How is fine-tuning billed differently? Managed podstack train jobs bill per run. Cap them with --budget <usd>.

Do data transfers cost extra? There’s no separate per-GB charge; transfers happen during the instance runtime you’re already paying for.

Next steps