---
title: GPU spend visibility
description: Read-only GPU cost commands in argusctl (cost month, now, forecast, idle) and how the Operations Console reports monthly burn, including the orphan-pod caveat.
---

# GPU spend visibility



ARGUS tracks GPU spend from append-only cost snapshots plus the live pricing catalog. All cost surfaces are **read-only**: they report and project spend, they never mutate anything. They are analytics-grade reads, so a PAT without any write scope can use them.

## The four cost views

### `argusctl cost month`

Month-to-date GPU spend, the current burn rate, and a breakdown by provider and tenant. This is the "what have we spent so far this month" view.

```bash
argusctl cost month
```

### `argusctl cost now`

The current instantaneous hourly burn, split into allocated and unallocated. Use it when you want a single number for "what is running costing right now."

```bash
argusctl cost now
```

### `argusctl cost forecast`

A straight-line run-rate projection from the current burn: hourly, then daily, then 30-day. It assumes nothing changes; it is a projection of the present, not a model of the future.

```bash
argusctl cost forecast
```

### `argusctl cost idle`

The actionable one: running pods that are serving zero cameras, and the dollars per day and per month each one burns. The output is effectively a "stop these" list. It supports `--watch` for a live-refreshing view.

```bash
argusctl cost idle
argusctl cost idle --watch 10
```

To act on the list, see [Stop, resume, or terminate a pod](/docs/compute/stop-resume-terminate).

## In the console

The COMPUTE page (route `#/compute`) shows monthly spend alongside the running pods and worker table. One honest-accounting caveat is built in: the monthly total only sums **tracked** workers. A running **orphan** pod (one that exists at the provider with no worker row) bills but shows as zero in the total, so the console renders a warning that burn is understated whenever running orphans exist. The orphan's actual live rate is visible in its pod drawer, which reads the rate straight from the provider.

## Where the numbers come from

* Compute spend comes from `argus.cost_snapshots`, an append-only table.
* Rates and projections lean on the GPU pricing catalog (see [Planner decisions and the pricing catalog](/docs/compute/planner-catalog)).

## Verify and cross-check

* `argusctl compute strain` shows the same idle pods from the utilization side (idle burn plus a recommendation).
* `argusctl compute list` confirms each worker's $/hr.

Command reference: [argusctl cost](/docs/cli/cost).


---

For a semantic overview of all documentation, see [/sitemap.md](/sitemap.md)

For an index of all available documentation, see [/llms.txt](/llms.txt)

For agent-facing discovery, including API and MCP surfaces, see [/agents.md](/agents.md)