Estimate what a GPU job will cost

The calculator below prices a job from the cheapest VM you could launch on DeployGPU right now, so the estimate is what you would pay if you started it today. Pick a GPU, the number of GPUs in the VM and the hours the job runs. It shows the compute cost, the balance you need to launch, and what a card top-up for the run would charge, fee included.

How the estimate is worked out

Compute is the hourly price of the offer times the hours you enter. DeployGPU settles paid time to the second: launching reserves the first hour from your balance, billing continues in hourly blocks, and when you stop, the unused seconds of the current hour go back to your balance. A 2.5-hour job therefore costs 2.5 times the hourly price, not 3. A launch that fails is refunded in full.

The top-up figure is the credit you would buy for the run, with the card processing fee on top. Card top-ups go through Stripe in amounts from $5 to $25,000, and each payment carries a fee of 5% plus $0.30, rounded to the cent and shown as its own line at checkout:

compute     = hourly price x hours
top-up      = compute, rounded up to the cent, at least $5 and at least one hour
card charge = top-up + (top-up x 5% + $0.30), per payment of up to $25,000

The hourly prices include any platform fee, so there is no other line to add for compute. The pricing page explains each charge in full.

What the estimate leaves out

The calculator covers compute on one VM. It does not include:

• Extra disk. On Verda you choose the disk size at launch, and space above the offer's included disk is added to the hourly rate. The deploy page shows that figure before you pay.
• Volumes on Verda or Hyperstack, which are billed every hour they exist, attached or not.
• Vast.ai containers, which are not in these prices, and Vast.ai bandwidth, which is billed separately.
• Price changes. Offers come and go and prices move, so a price here can be gone by the time you launch. The deploy page always shows the current breakdown before anything is charged.

When the balance runs out mid-job

No job is stopped at the number of hours you typed in. A DeployGPU instance runs until you stop it, with no idle detection and no automatic timer, so a forgotten VM keeps billing. When your balance cannot cover the next hour, the instance switches to per-minute billing until the balance reaches zero, and then it is stopped and its disk deleted. Emails go out when about 2 hours of runtime are left, when per-minute billing starts and when the instance is stopped.

The dashboard shows an estimated time left: your balance divided by the hourly rate of everything you have running. Auto top-up on the Billing page refills the balance at a threshold you choose. For scripted jobs, API cost guardrails shows how to stop machines automatically, and the stop guide covers what to save before you stop.

Choosing the GPU before you price it

Price per hour is only half of the cost: a faster GPU that finishes in half the time can cost less overall, and a model that does not fit in memory will not run at any price. The GPU overview compares the cards, how much VRAM fine-tuning needs sizes training jobs, and the KV cache guide sizes inference. For anything bigger than one 8-GPU VM or a fixed term, request a cluster quote.