Datacenter GPU hosting vs independent hosts is a choice you make with one switch on the Cloudj GPU deploy page. Every machine in the results is an independent host with its own owner, and some of them carry a Datacenter badge because the host sits in a data centre. Turn on Datacenter only and the list keeps just those. This article explains what the badge does and does not tell you, when the filter is worth the shorter list, and why the backup rules stay the same whichever you choose.
For the bigger picture of how location and filters work together, see the GPU cloud by location guide.
What the badge means
The Datacenter badge describes where the host machine is located: in a data centre rather than, for example, a smaller private facility. It is a property of the host, not of the GPU model. A consumer RTX card can sit in a data centre, and a datacenter-class card such as an A100 is not automatically shown with the badge.
What the badge does not mean is just as important:
- It is not an SLA. The console's own wording applies to every machine: "Machines are independent hosts with no hardware SLA. Keep backups of important data."
- It does not add redundancy or an automatic move. If any host goes offline, billing pauses while it is offline, and there is no automatic move to another machine.
- It does not change billing rules. Per-minute billing, the deposit, stop and destroy work the same way.
- It does not reveal a city or facility name. Location remains country level only.
What the filter changes in practice
Turning on Datacenter only changes the list you choose from, not the product you get.
| All hosts | Datacenter only | |
|---|---|---|
| Number of listings | More | Fewer |
| Typical price position | Wider range, including lower prices | Often higher |
| Chance of finding a specific model in a given country | Higher | Lower |
| Host environment | Mixed | Data centre facilities |
| Hardware SLA | None | None |
| Backups needed | Yes | Yes |
Data centre facilities generally have professional power, cooling and connectivity, which many teams prefer for services that run for days or weeks. That preference is reasonable, but it is a preference about the environment, not a guarantee about your instance.
When to turn the filter on
The filter tends to make sense when:
- Your policy or client asks for it. Some contracts or security questionnaires ask whether processing happens in a data centre. The filter lets you answer that question for the hosts you choose; check what else the questionnaire requires.
- You run a long-lived service. For an inference API that should stay up for weeks, a host in a data centre may be your preference, alongside a high reliability score and a long max duration.
- You need predictable network quality. Combine the filter with minimum download speed and maximum bandwidth price.
The reliability score guide explains how to read the reliability percentage, which matters at least as much as the badge.
When to leave it off
Leave the filter off when:
- The job is short or restartable. Experiments, evaluation runs and batch jobs with checkpoints do not need a specific facility type.
- Price matters most. A wider list gives the price sort more to choose from.
- The country list is narrow. If you must stay in one or two countries, adding Datacenter only may leave you with nothing. Search without it first, then decide.
- You need a specific model now. A rarer model in a specific country is more likely to appear without the filter.
A quick decision path
- Open the GPU deploy page, set your country codes and model.
- Search without Datacenter only and note what is listed.
- Turn the filter on and compare. If acceptable rows remain, decide whether the difference in choice and price is worth it for this job.
- Whatever you pick, check reliability, max duration and the "available until" date.
Stock changes during the day, so the comparison can come out differently tomorrow. Live prices are on GPU cloud.
Model class is a separate question
Do not confuse the Datacenter badge with datacenter GPUs. Whether you need an A100 or an RTX 4090 is a question about memory, features and account level. Datacenter and workstation GPUs need a higher account level than new accounts have. The consumer vs datacenter GPU comparison covers that decision.
Backups either way
Because every listing is an independent host with no hardware SLA, the same backup habits apply on both sides of the switch:
- Copy checkpoints and outputs off the machine regularly.
- Keep weights and configuration where you can redeploy quickly.
- Plan around the "available until" date; the instance and its disk are removed when it arrives.
- Use a stable VPS in front of public services so you can switch GPU hosts without changing DNS.
The guide to backing up a GPU instance before its availability ends gives a step-by-step routine.
FAQ
Does the Datacenter badge mean an SLA?
No. Every machine is an independent host with no hardware SLA. Keep backups.
Are datacenter-located hosts always more expensive?
Not always, but they are often higher in the price sort. Compare both views in the console.
Is an A100 always shown with the Datacenter badge?
No. The badge describes where the host is, not the GPU model.
Can I combine Datacenter only with a country filter?
Yes. The filters work together; expect fewer results.
Next step
Open the GPU deploy page, run your search with and without Datacenter only, and pick the row that fits this job. Whichever you choose, keep backups and plan around the availability date.