A GPU cloud North America search on Cloudj starts with two country codes, US and CA, typed into the Country field of the GPU deploy page. There is no "North America region" to pick and no Cloudj data centre behind the list. Instead you see a live inventory of independent host machines, each reporting the country it sits in, and you choose the one that fits your job. This guide explains how to run that search well, what the results tell you, and when a North American GPU is the right call even if your users live in Asia or Europe.
If you want the general picture of how location works for every region first, start with the GPU cloud by location guide. This article stays on the Americas.
How a GPU cloud North America search works
Open the GPU deploy page and enter US, CA in the Country field. Codes can be separated by commas or spaces, and you can mix in any other two-letter code. The list then shows machines whose host reports one of those countries, sorted by total price with the lowest first, up to 50 machines.
Location is reported at country level only. A result tagged US tells you the machine is in the United States, not which state or city. If your plan depends on being in a particular city, a GPU listing cannot promise that; a Cloudj VPS can, which we come back to below.
Each row gives you what you need to compare hosts:
- GPU model and count (1x, 2x, 4x or 8x, all in the same machine)
- VRAM, vCPU, RAM and download speed
- reliability, shown as a percentage
- country
- max duration in days, plus an "available until" date
- storage price per GB-month and bandwidth price per TB down and up
- a Datacenter badge when the host sits in a data centre
The filter walkthrough covers every field in detail.
Why North America is often the first place to look
In practice, the United States often shows the widest spread of models in the list, from consumer GeForce cards to datacenter A100 and H100 machines. That is a general pattern, not a promise: stock changes during the day, and a model you saw this morning may be rented by the afternoon. Treat the US as a good starting point for breadth, then check the console for what is actually listed.
Canada is worth adding to almost every North American search. Typing US, CA instead of US costs nothing and widens the pool. For teams with users in Canada, a CA host may also sit closer to them. The dedicated guides for the United States and Canada go deeper on each country.
Matching the search to your workload
Not every job needs a GPU near the people who use it. The useful question is whether a human is waiting on each response.
| Workload | Does location matter? | Suggested Country field |
|---|---|---|
| Chatbot or image API for North American users | Yes, keep round trips short | US, CA |
| Fine-tuning or training run | Rarely; data transfer matters more | US, CA, plus any region with the model you need |
| Batch rendering or transcoding | No, as long as uploads are reasonable | Leave wide, sort by price |
| Interactive notebook for a team in Asia | Yes, for comfort | An Asian code first, US as fallback |
| Work that must stay inside one country | Yes, by policy | Only the allowed code |
When a US GPU suits users in Asia or Europe
A team in Vietnam, Japan or Germany can still be well served by a North American machine. Batch work is the clearest case. A fine-tuning job that runs for hours does not care about the extra distance once the dataset is uploaded. A rendering queue, a nightly embedding job or an evaluation sweep behaves the same way.
The trade-off sits in data movement. Large datasets and model weights take longer to cross an ocean, and every gigabyte you download or upload is charged at the machine's listed bandwidth price. Before choosing a distant host for batch work, check two numbers in the row: the bandwidth price and the download speed. A cheap GPU with expensive bandwidth can cost more overall than a slightly pricier host with cheap transfer.
For interactive work, such as a Jupyter notebook or a live chatbot, distance is felt on every request. If a team in Asia needs to iterate quickly, look at the Asia guide first and keep US as a fallback.
Pairing a GPU with a North American VPS
GPU hosts come and go. A machine shows an "available until" date, and when that date arrives the instance and its disk are removed. If you serve an API, you do not want your domain tied to a GPU that may change next month.
The usual pattern is a small, stable VPS that holds your domain, TLS certificate, authentication and request queue, with the GPU behind it. Cloudj runs VPS in the United States in eight cities (New York, Los Angeles, Santa Clara, Seattle, Chicago, Dallas, Atlanta and Miami) and a VPS in Canada in Toronto. Because the VPS city is fixed, you can place the front end near your users even though the GPU behind it is only known at country level. When you swap GPUs, you change the tunnel target on the VPS and DNS stays the same.
For a broader look at the Cloudj VPS locations on this side of the world, see the Americas, Oceania and Israel VPS guide.
When nothing is listed
Sometimes a search for a specific model with US, CA returns no machines. That is normal for a live inventory. Work through these steps:
- Loosen the filters. Lower minimum reliability a little, raise the maximum price per hour, or reduce minimum VRAM if your model fits in less.
- Try a sibling model. If one card is not listed, a card with the same or more VRAM often does the same job.
- Add more codes. For batch work, add European codes such as
DEorNL; the extra distance rarely matters. - Check again later. Stock changes during the day, so a model missing now may appear within hours.
When Cloudj is not the right fit
Be honest with yourself about requirements. Machines are independent hosts with no hardware SLA, and there is no automatic move to another machine if a host goes offline. If your contract requires a guaranteed machine in one specific US state, or a written SLA, a GPU from a live inventory is the wrong tool. For most experiments, inference services with a VPS front end and batch jobs with backups, it fits well.
New accounts start at a level that covers consumer GPUs. Datacenter cards such as A100 and H100 need a higher account level; the console tells you when a choice is not available and you can contact Cloudj support from there.
FAQ
Can I choose a US state or city for a GPU?
No. GPU location is shown at country level only. If you need a fixed city, put a Cloudj VPS in that city in front of the GPU.
Should I search US and CA together?
Usually yes. Adding CA widens the list at no cost, and the country is still shown on each row so you can choose.
Is a US GPU a good idea for a team in Asia?
For batch jobs such as training, rendering or transcoding, often yes. For interactive inference or notebooks, an Asian host usually feels better.
How do I see prices?
Prices change every few minutes. See live "from" prices on GPU cloud; the console shows the exact price, which is locked once you deploy.
Start your search
Open the GPU deploy page, enter US, CA in the Country field and pick the model you need. Compare reliability, max duration and bandwidth price before you deploy, and keep backups of anything important. You can browse every model and its live price on GPU cloud first.