Cost Explorer

Analyze, forecast, and optimize GPU spend across your organization

Know exactly where your GPU dollars go. The Cost Explorer gives you full visibility into spend across teams, clusters, GPU types, and workload classes — with built-in forecasting, waste detection, and optimization recommendations to help you get more out of every GPU-hour.

Navigate to Cost Explorer in the Chamber dashboard sidebar to get started.

At a Glance

Cost Explorer is built to answer the questions every GPU infrastructure team asks:

• How much are we spending? Total and per-team GPU costs with trend analysis
• Where is the money going? Breakdown by team, user, cluster, GPU type, and workload class
• What will we spend next? Month-end forecasts with confidence intervals
• What are we wasting? Idle capacity, failed workload costs, and underutilized resources
• How do we optimize? Actionable recommendations for spot instances, rightsizing, and failure reduction

Key Metrics

Four headline numbers are displayed at the top of every Cost Explorer view:

Metric	Description
Total Spend	GPU cost from completed workloads in the selected time range
Month to Date	Cumulative GPU spend so far this calendar month
Projected Month-End	Where you’ll land by end of month, based on current trajectory
Trend vs. Prior	Percentage change comparing the recent period to the prior period

Filtering and Grouping

Filters

Slice your data by any combination of:

• Team — One or more teams
• Cluster — Specific capacity pools
• GPU Type — Hardware type (A100, H100, etc.)
• Workload Class — Type of workload (training, inference, interactive, etc.)

Filters apply across every chart, metric, and table on the page. Active filters appear as removable pills below the filter bar.

Grouping

Change how the spend chart and cost breakdown are organized:

• Team / User / Cluster / GPU Type / Workload Class

Switch grouping at any time to see costs from a completely different angle.

Time Range

Pick from presets or define your own window:

• Last 7 / 30 / 90 days
• This month / Last month
• Custom range with a calendar picker

Filters and time range are saved in the URL — bookmark or share a specific cost view with your team.

Spend Over Time

A time-series chart shows how your spend trends across the selected period:

• Granularity — Toggle between daily, weekly, or monthly aggregation
• Stacked series — Top contributors are shown individually; smaller ones grouped as “Other”
• Forecast overlay — Project costs forward with a shaded 95% confidence band

Forecasting

Turn on the forecast overlay to see where spend is heading:

• Uses linear regression fitted to your daily historical data
• Shows a projected line with upper and lower confidence bounds
• Defaults to an end-of-month projection, but you can pick any date up to 365 days ahead
• Trend detection automatically classifies your spend as increasing (over 10% growth), decreasing (over 10% decline), or stable

Capacity Utilization

This section compares what you’re paying for (provisioned GPU capacity) against what’s actually being used by workloads:

Metric	Description
Total Capacity Cost	Cost of all provisioned GPUs across selected clusters
Used Cost	Cost attributable to completed workloads
Wasted Cost	Idle capacity — provisioned but sitting unused
Utilization %	Percentage of provisioned capacity actively used

Utilization is color-coded so you can spot problems at a glance:

• Green (70%+) — Healthy
• Orange (40-70%) — Worth reviewing
• Red (under 40%) — Significant optimization opportunity

Capacity metrics come from cluster telemetry. Team and Workload Class filters only affect the “Used Cost” number, since provisioned capacity isn’t tied to specific workloads.

Cost Breakdown

A donut chart shows the proportional distribution of spend by your selected dimension. Items contributing less than 1% are grouped as “Other” — hover to expand the full list.

Cost Events Table

Dig into the details with a full table of individual cost events:

Column	Description
Date	When the cost was incurred
Team	Team name
User	Who submitted the workload
Cluster	Capacity pool
GPU Type	Hardware type
Workload Class	Type of workload
Hours	GPU instance-hours consumed
Rate	Hourly rate for this GPU type
Cost	Total cost for this event

Pagination and Export

• Navigate results with cursor-based pagination
• Set your page size: 10, 25, 50, or 100 rows
• Export to CSV to pull data into spreadsheets or external tools (up to 100K rows)

GPU Rate Configuration

Cost calculations rely on per-GPU-type hourly rates. Org Admins can set these up:

1. Click the settings icon in the Cost Explorer header
2. Set the hourly rate for each GPU type in your organization
3. Rates apply retroactively to all cost calculations

Waste Analysis

Chamber automatically surfaces wasted GPU spend from three sources:

Waste Category	Description
Failed Workloads	GPU time consumed by workloads that ultimately failed
Preempted Workloads	GPU time from workloads preempted before completion
Short Runtime, High Cost	Workloads under 5 minutes that cost more than $1 — often a sign of misconfiguration

Optimization Recommendations

Chamber analyzes your usage patterns and surfaces actionable next steps:

Spot Instance Savings

Identifies workloads that could run on spot or preemptible instances, with estimated monthly savings.

Failure Reduction

Flags teams or GPU types with failure rates above 20% and calculates the cost impact.