# Manage data center hosts

> See your enrolled hosts, GPU measurements, and idle-capacity lending in one place.

Source: https://www.nodus-compute.ai/docs/guides/pools/data-center/
Build revision: 4c89a51d7dcfd856f1373531796fbb15831d8da3

**Beta:** this feature may change.

Sign in with your Nodus account and open **Data center**. Select your organization, project and pool. The page uses your existing organization permissions. It shows enrolled hosts belonging to that project and pool; **All organization hosts** opens the broader host list you are authorized to read.

Choose **Create a pool** and review its routing fee, then **Enroll a host** to connect a machine. Follow the [host prerequisites and enrollment guide](https://www.nodus-compute.ai/docs/guides/pools/). Inventory appears after the agent reports it. A missing or stale reading appears as unavailable, rather than zero utilization. Host utilization is the mean across its GPUs. It uses the GPU sample time and expires after 30 seconds; a newer heartbeat alone does not refresh that reading. A measured idle reading shows 0%.

The hourly activity chart shows recorded busy and idle device-hours; missing hours remain blank. GPU utilization while busy is a separate reading. Unmeasured online time means missing utilization readings and can overlap busy or idle occupancy. One device-hour is one GPU for one hour. Choose a time window and **Refresh measurements** for a new report. The report fetch time is separate from the hourly measurement times, which use UTC.

Under **Lend idle capacity**, review the published agreement and each host’s eligibility before choosing Yes. Idle measurements do not establish that a host can accept outside work or that it will earn money. Lending eligibility, active borrowing and cleanup status come from the sharing service. Cash earnings are separate from Nodus credits; cash payouts require an eligible account and accepted cash terms. See [lending](https://www.nodus-compute.ai/docs/guides/pools/liquid-capacity/) and [cash payouts](https://www.nodus-compute.ai/docs/guides/pools/cash-payouts/).

**Ready** describes node health and can appear while participation is disabled. Running capacity work also requires current operator consent, host eligibility and local allocation permission. Before returning a physical host to a customer, complete the host package’s retirement and cleanup procedure. The portal is not a reimaging or bare-metal handoff service, and unenrollment alone is not a clean-device proof.

The [open-source host package](https://github.com/nodus-compute/nodus-dc/releases/tag/v0.1.0-dev8) also works independently, with its own local dashboard. Installing it does not grant lending permission. Follow its published support matrix and connection instructions; the hosted portal does not remotely partition GPUs.