The simplest way to rent GPU USA capacity on Cloudj is to type US into the Country field of the GPU deploy page and let the console list the machines that report the United States as their country. You are not choosing a Cloudj region or a city. Each row is an independent host machine with its own GPU model, reliability score, maximum rental duration and bandwidth price. This guide shows how to read that list, which models people usually look for in the US, and how to keep a stable API in front of a GPU that may change over time.
If you are comparing the US with Canada or deciding whether North America suits you at all, the North America GPU cloud guide is the better starting point.
What "US" means in the results
Location on Cloudj GPUs is reported at country level only. A row tagged US tells you the host is in the United States. It does not tell you the state, the city or the data centre, and there is no way to filter by them. If you need to be near New York rather than Los Angeles, a GPU listing cannot promise that. A Cloudj VPS can, as you will see below.
That limitation matters less than it sounds for most GPU work. Training, fine-tuning, rendering and batch inference rarely care which coast a machine is on. Interactive services care more, and for those the stable VPS in a known city does the heavy lifting on latency to users.
How to rent GPU in the USA, step by step
- Open the GPU deploy page, which preselects the RTX 4090. Change the model if you need another.
- Enter
USin the Country field. - Set a minimum reliability you are comfortable with and, if you serve traffic, a maximum bandwidth price per TB for downloads and uploads.
- Set the disk size you will need. Prices in the list use that size, and disk cannot be resized after deploy.
- Read the results, sorted by total price, lowest first.
- Pick a machine, choose a template and at least one SSH key, and deploy.
The country filter walkthrough covers every field in more depth.
Models people look for in the US
The United States often shows a broad spread of models, from consumer GeForce cards to workstation and datacenter GPUs. That is a general observation about breadth, not a statement that any model is in stock right now. Search the console when you need one.
| Model | VRAM | Typical use |
|---|---|---|
| RTX 4090 | 24 GB | Inference of quantized LLMs, image generation, LoRA training |
| RTX 5090 | 32 GB | Same jobs with more headroom for larger models or batches |
| A100 PCIe | 40 GB version | Training and batch work that benefits from datacenter features |
| H100 SXM | 80 GB | Larger training runs and high-throughput serving |
Datacenter and workstation GPUs such as A100 and H100 need a higher account level than a new account has. The console says so when a choice is not available; contact Cloudj support from there if you need it. Live "from" prices for every model are on GPU cloud.
A worked example: two US listings side by side
Say you search for RTX 4090 with US and two rows catch your eye. Here is how to compare them without any price numbers:
- Machine A has a higher reliability percentage, a long max duration and a moderate bandwidth price.
- Machine B is a little lower in the list on price, has a shorter max duration and a cheaper GPU rate, but a higher bandwidth price per TB.
If you are serving an image API that sends many large files out, Machine A will likely cost less overall because outbound transfer is charged per GB at the listed rate. If you are running a two-day fine-tune with a small dataset, Machine B may be the better pick, as long as its "available until" date covers the run with a margin. The total price column already includes GPU and disk; bandwidth is on top and depends on how much data you move.
Use the bandwidth filter on purpose
Many first-time users ignore the bandwidth columns. For US machines that serve users abroad, or that pull large datasets, bandwidth can become a large part of the bill. Two habits help:
- Set a maximum bandwidth price for downloads when you plan to pull big model weights or datasets.
- Set a maximum for uploads when your service sends images, audio or video back to users.
You can always widen these limits if the list becomes too short.
Pair the GPU with a US VPS
Cloudj runs VPS in the United States in eight cities: New York, Los Angeles, Santa Clara, Seattle, Chicago, Dallas, Atlanta and Miami. That makes the VPS the right place to control where your users connect. Put the domain, TLS certificate, authentication and request queue on a VPS in the city closest to most users, and forward inference requests to the GPU over an SSH tunnel. When the GPU host's "available until" date approaches, deploy a new GPU, point the tunnel at it, and your users never see a change.
If you are unsure which US city to pick, the US East vs West VPS comparison is a practical guide.
When nothing is listed
Even a broad country returns empty results sometimes, especially for one specific model with strict filters. Try these, in order:
- Relax reliability or price limits a little.
- Try a sibling model with the same or more VRAM.
- Add
CA, or for batch work add European codes such asDEandNL. The US vs Europe GPU cloud comparison helps you decide when Europe is fine. - Check again later; stock changes during the day.
For long training jobs on US hosts, the AI training on US GPUs guide covers checkpoints and data transfer.
FAQ
Can I choose a US state for my GPU?
No. GPU location is country level only. Choose the city on your VPS front end instead.
Are US GPUs billed differently?
No. Every GPU is billed per minute from your account balance: GPU plus disk while running, storage only while stopped, and bandwidth per GB.
What if a host goes offline?
Billing pauses while it is offline, but there is no automatic move to another machine. Keep backups so you can redeploy elsewhere.
Do I get a dedicated IP and port 22?
The Connection tab shows the host, the SSH port and the full command. The port is not always 22.
Ready to look
Open the GPU deploy page, set Country to US and compare a few rows on reliability, max duration and bandwidth price. Keep your domain on a US VPS, and keep backups so you can move between hosts whenever you need.