Rent an RTX 5090 in the Cloud by the Minute (2026)

You can rent an RTX 5090 (32 GB of GDDR7) on cloudgpu.app by the minute: pick it on the deploy page, and the machine is ready in about 2 minutes with root SSH. You pay with USDT, Alipay, Google Pay, Apple Pay or a card, so a US credit card is not needed, and the current rate is on the pricing page.

What it costs

Billing is per minute, rounded up, with a 1-minute minimum. The rate for each card is on the pricing page; it changes with supply, so this post does not repeat it. When you deploy, one hour is pre-authorised from your balance, and unused minutes go back to your balance when you destroy the instance. Rules and worked examples: how per-minute billing works.

Why per-minute billing matters for short jobs

Many GPU jobs are short: a first test of a model, a batch of images, an evaluation run. With per-minute billing you pay for the minutes you use instead of rounding every session up to a full hour, and that rounding adds up when you run several short sessions a day.

Paying without a US credit card

Top up with USDT (TRC-20) on the billing page, or use the card checkout, which takes Alipay, Google Pay, Apple Pay, Visa and Mastercard. There is no subscription, and unused balance does not expire. Minimums and details are on the payments page and in how to pay for GPU rental without a US card.

What you get with the machine

• Root SSH access with a per-instance password.
• A template from the deploy page. Root access means you can install anything else yourself.
• For templates that expose an API, such as Ollama and ComfyUI, a public HTTPS endpoint of the form https://<random-id>-i.cloudgpu.app, reachable only while the instance runs. See Run Ollama on a cloud GPU with a public API.
• JupyterLab for file management.

What an RTX 5090 is for, and when to pick another card

The RTX 5090 is a Blackwell card with 32 GB of VRAM. People rent it to run open models, for LoRA and QLoRA fine-tuning, and for image and video generation. VRAM need depends on model, precision and workflow, so check yours first: how much VRAM an LLM needs for language models, VRAM for FLUX and SDXL for image models. For specs side by side, see RTX 4090 vs RTX 5090 vs A100.

Pick another card when:

• You need more than 32 GB on one card: the deploy page lists the larger VRAM classes.
• You need the machine in a specific jurisdiction: check before putting regulated data on it. See /pricing for available locations.

Rules of the house

Cryptocurrency mining and anything illegal get the instance destroyed and the account banned. Inference, fine-tuning, image and video generation, notebooks and evaluation runs are all fine.

How to start

1. Sign up at cloudgpu.app with email or Google.
2. Top up on the billing page.
3. On the deploy page, pick a template and RTX 5090, then click Deploy.
4. After about 2 minutes, the SSH details and, for API templates, the HTTPS endpoint are on the dashboard.
5. Save your outputs, then destroy the instance. Unused minutes go back to your balance.

If a card you need is missing from the list, tell us on the contact page.

Frequently asked questions

How much does an RTX 5090 cost on cloudgpu.app?

The current rate is on the pricing page. Billing is per minute, rounded up, with a 1-minute minimum.

Is there a minimum rental time?

One minute. One hour is pre-authorised from your balance when the machine starts, and unused minutes go back to your balance when you destroy the instance.

Can I pay without a credit card?

Yes. Top up with USDT on the TRC-20 network, or use the checkout with Alipay, Google Pay or Apple Pay. There is no subscription.

How long does it take to get a machine?

About 2 minutes from clicking Deploy.