Launching today
CloudGPU
Click Deploy, make a coffee, come back to a rented GPU
1 follower
Click Deploy, make a coffee, come back to a rented GPU
1 follower
GPU rental by the minute. Model APIs on the same balance. One prepaid balance; card, Alipay, Google/Apple Pay or USDT from $5. No phone number or US card needed.
Hi Product Hunt, I'm Yang, the maker of CloudGPU.
Here's the picture we built CloudGPU around. You pick a template and click Deploy. You go make a coffee. In about 2 minutes there's a browser URL waiting, with root SSH, and a public HTTPS endpoint if you picked Ollama or ComfyUI.
You run your experiment for 7 minutes. You pay for 7. GPUs are billed per minute with a 1-minute minimum; each started hour is pre-charged, and when you destroy the instance the unused minutes go back to your balance.
That same balance also pays for model APIs (chat, image, speech, transcription) behind one OpenAI-compatible endpoint. Top up once, spend on either.
Paying shouldn't be the hard part either: sign up with an email or a Google account, no phone number needed. Top up with Alipay, Google Pay, Apple Pay, card or USDT (TRC-20). No subscription, and the balance doesn't expire.
Honest notes: we're in public beta with no SLA, and prices and stock change, so cloudgpu.app/pricing and cloudgpu.app/deploy are the live source.
What would you run first?
One real deploy, start to finish:
1. On Deploy, pick the Ollama template and a card with enough VRAM for your model (4090, 5090, A100 and more).
2. About 2 minutes later, the instance row shows SSH plus an API endpoint like https://<random-id>-i.clou.... Pull any model with `ollama pull` over SSH (some template versions come with a small one pre-pulled).
3. Ollama speaks OpenAI on /v1, so any OpenAI SDK works:
client = OpenAI(base_url="https://<random-id>-i.clou...", api_key="ollama")
Prefer not to host the model? Our hosted OpenAI-compatible gateway is covered in the API docs at cloudgpu.app/api/docs.
Click Destroy when you're done.
What would you run first?
ā Yang