/portfolio route) is your central view of every GPU reservation your organization holds: reserved windows from Reserve and the Market, and running VMs from Spot.
A reserved booking does not launch compute automatically. Launch VM or Bare Metal from its detail page, or select its nodes for Kubernetes or Slurm from Orchestration. A filled spot bid starts its VM for you.
Spot reservations are listed with your other reservations and run until they end. Stop VM ends a spot reservation, and Ornn credits the unused part of the current hour. See How spot ends.
The Active, Upcoming, Offers & listings, and History tabs are hidden only when all four sections are empty. If any section has data, the tabs remain available.
Nodes tabs
While reservations load, Nodes keeps the breadcrumb visible and waits to show the collection heading, tabs, and rows until their state is known. Previously loaded results remain visible during background updates. Once Nodes has records, the heading and tabs remain visible on the first visit to another section, even while that section loads. Switching between reservation tabs reuses recently loaded reservations, and returning to Offers & listings keeps the last offers visible while live updates reconnect. If you have no reservations, it shows a centered illustration and a Browse reservations action. The tabs remain available when another section has records, and search appears when the current tab has records to browse.
Your open spot bids are on the Spot page.
The row’s three-dot menu includes Relist and Reboot. Actions are disabled
while availability is checked or when the reservation or its nodes are not eligible.
Use the search field below the tabs to filter the loaded reservations in the
current tab. Search matches displayed details and keeps grouped bookings
together. Clear the search to restore the full list. View changes row spacing
without removing columns or changing your reservations.
Access setup is not a Nodes tab. You configure VM or Bare Metal from the reservation detail
page itself. See Choose how to access your Ornn compute.
What each reservation row shows
Each row in Active, Upcoming, and History displays:- Type: the GPU label, for example
H100 SXM. - Region: the facility location. If a node’s IP has no location match, the row keeps the listing’s facility. Locations without a recognized flag align directly with the column heading, with no empty icon space.
- Quantity: VM rows show the purchased GPU count, for example
1 GPUor4 GPUs. Bare Metal rows show the reserved GPU count and booked host count, for example64x (8 nodes). - Term: for a reserved window, its full start and end. A spot reservation shows when it started. Scroll the table horizontally on narrower screens to see all columns.
- Price: the price per GPU-hour you pay.
/portfolio/[reservationId].
A reservation ownership change moves billing and permission to manage the reservation to the new
organization. It leaves machine data, running workloads, and installed credentials unchanged. An
ownership change does not erase data or prepare the machine for a new customer.
Reservation statuses
Reservations scheduled for later, or awaiting node assignment show a pending status message on the detail page. This view does not include Terms, Invoices, or Relist sections. Rows with assigned machines show a health dot. Rows without machines have no dot or empty space before the GPU name. Reservation status describes your booking. Cluster or machine status separately describes the environment you launched.Reservation detail page
The full detail page at/portfolio/[reservationId] is where you manage everything tied to a single reservation: configure access, review connection details, and monitor utilization.
Select the reservation in the header to search and switch to another booking.
The menu shows quantities, dates, and reservation IDs to help distinguish bookings.
In a short browser window, scroll the reservation page to reach the node controls below its summary and notices.
The Nodes panel lists every booked host. Hosts that have joined Kubernetes or Slurm appear in a numbered cluster group even when they are no longer listed as individual Bare Metal machines. Open the group to inspect each host.
Hosts from cancelled, completed, or expired reservations do not appear as live workers on a related active reservation’s Nodes panel, even if the reservations shared a cluster.
Bookings with different facilities, terms, or rates stay separate. Sharing a cluster does not add another booking’s hosts to this page’s Nodes panel or node actions.
If a background refresh fails, the page keeps its last loaded nodes, connection
details, and charts visible.
Your selections and available node actions are preserved.
The reservation Observability tab has separate Metrics and Logs views for standalone VM and Bare Metal hosts. Switch views at the top; filters are retained and the hidden view pauses its data requests. After hosts join Kubernetes or Slurm, open the cluster on Orchestration and use its Observability tab.
The reservation dashboard opens with four GPU charts. Choose System metrics for CPU, memory, storage, and network latency, or choose an individual metric below the toolbar. Use the Last 1 hour menu to choose a shared metrics window. Use Pause live to pause updates. Panning or zooming also pauses updates so you can inspect history. Return to live resumes updates at the current time. Console logs have their own time controls. Logs open on the last hour, with the newest entries at the bottom. Search the loaded entries and filter by severity or host. Column headings stay above the scrolling log entries. Long messages and node addresses wrap within their columns. On small screens, each entry shows its node address and date above the message. Choose Load older logs to search further back within the selected window. Pause live logs stops incoming entries; Resume live logs returns to the current time.
Metric charts open on the past hour. Scroll over a chart to zoom in or out, and drag to move through time. Earlier history loads as you explore, and live updates continue when you are viewing the current time.
Open Hosts: All to search and select hosts using checkboxes. Keep at least one host selected, or choose All hosts to restore the full selection. The selection applies to every chart. Each host keeps the same line color across charts, with labels in each chart legend. Choose GPU metrics, System metrics, or an individual metric from the visible controls below the toolbar. Your host selection and metrics time range stay in place when switching views.
For a single host, the graph color matches the selected metric. When comparing multiple hosts, each line matches its host selector color and keeps that color when you switch metrics or isolate a host.
If history fails to load for any selected host, an existing chart keeps its
previously loaded data.
Dashboard metrics are restricted to your organization’s authorized reservations
and selected hosts. Unsupported metric queries do not return telemetry.
For authorized physical nodes, the telemetry API also provides optional per-GPU
readings in the JSON-encoded gpus field. Select a GPU by UUID with gpu_uuid
on history queries, spanning up to 24 hours. These reads retain reservation time
boundaries.
Charts show Node metrics by default. For physical hosts reporting per-GPU
readings, isolate one host by its IP, then select one or more GPU buttons to
compare temperature, power, fan speed, utilization, and memory usage. Unavailable
fan readings show —. Select Node to return to the aggregate view.
GPU charts show up to 24 hours at a time; drag to explore earlier history.
A confirmed or active reserved window can be relisted. Choose nodes and a time slice inside the reservation, then set an asking price. Buyers buy windows of it at that price, and idle hours sell on Spot at it. Your relists and Delist are under Offers & listings. See Relist capacity.
Who uses the machines
When you sell a window of your reservation, the buyer uses those machines for that window and you get them back afterwards. When idle relisted time runs on Spot, the spot buyer uses the machine until the spot VM ends. If capacity is resold more than once, a machine goes to the last buyer in that chain of resales whose window covers the current time. Use Download specs in the reservation header’s ellipsis menu to save a PDF of the listing specifications. You can download specifications for a private reservation or after its listing is withdrawn or expires. The download requires you to be signed in to the organization that owns the reservation. For Bare Metal, this menu also shows Switch public network and Reboot all nodes. Offline nodes can still be selected in the Nodes tab. Choose Relist in the selection toolbar to open a relist with the selected node count filled in. Unavailable actions stay visible but disabled. Reboot all nodes is available only when every node in the reservation is eligible to reboot; use the Nodes tab for eligible subsets.Access setup
If your reserved window has started but no nodes are assigned, the detail page shows Delivery pending. This also applies if all previously assigned nodes are removed. Machine access becomes available after nodes are attached; your reservation terms and billing remain available. VM or Bare Metal launch controls and SSH key attachment live on this page. Kubernetes and Slurm setup opens Orchestration; cluster membership does not turn these hosts into standalone SSH machines. The legacy/portfolio/access URL redirects here. See Choose how to access your Ornn compute for prerequisites and the full flow.
If Cluster status unavailable appears, the page could not load cluster
information. This warning alone does not mean your VM or Bare Metal machine has
stopped. Refresh the page to retry.
Connection details
Once access is launched, the detail page shows:- SSH Keys tab: Keys show their fingerprints; use the copy or download action to get the full public key. Use See all account keys to reveal saved keys you can add to this reservation.
- SSH key sync: Sync registered keys to each running or provisioning VM in the reservation.
- VM details: Click a VM row, or focus it and press Enter or Space, to open its detail side panel and copy its connection values. Checkboxes select VMs independently. The panel’s action menu lets you copy the SSH command or open the stop confirmation; stopping applies to all VMs in the reservation.
- SSH access: the SSH Host, User, and Port fields for connecting to your allocated machines.
- Quick Connect: a ready-to-run SSH command you can copy to your clipboard.
Utilization, Spend, and SLA Uptime
The detail page’s Utilization and SLA Uptime panels render a placeholder until real telemetry has been received; Wait for measured usage before interpreting them. For a reserved window, Spend accrues from the reserved start through the earlier of now or the reserved end, multiplied by GPU count and the per-GPU-hour price. A reservation that has started shows a live figure before any telemetry reports and before you launch. It is an estimate of cost accrued so far, not an invoice; invoices are in Billing. A cancelled reservation retains its scheduled dates but accrues no usage charges. Starting or stopping a VM during the reserved window does not change what you pay. For a reserved window, the Total commitment figure below the Spend value is the full cost across the window (GPU count × price × hours). Spot hours are prepaid hourly; see Spot hours.The reservation detail page reads GPU quantity, per-GPU-hour rate, and term from the booking. If
you booked a reserved term, the detail page shows those booked numbers rather than any internal
deploy defaults on the underlying compute rows.
When telemetry starts flowing, the Utilization panel reads
<used>/<total> GPU hours (for example, 120/960 GPU hours), with the total derived from GPU count and term length. A secondary line below the value shows the recent-jobs synced count and a staleness figure.CLI equivalents
Inspect reservations from the Ornn CLI:What’s next
Relist capacity
Sell windows of a reservation and put idle time on Spot.
Choose your access mode
Configure VM or Bare Metal access from the reservation detail page.