💻

GPU Utilization ROI Calculator

Find the real return on a rented or owned GPU: how idle time inflates your effective cost per productive hour, and whether the revenue it earns beats the bill.

GPU cost per hour

Hours paid per month

Utilization

Revenue per productive hour

Loading calculator…

How it works

Effective cost/hr = (cost/hr × hours) ÷ (hours × utilization%) · ROI = (revenue − cost) ÷ cost × 100

You pay for a GPU by the clock, but it only earns money while it is busy — so idle time silently inflates your true cost per unit of work. At 40% utilization a $2/hour GPU actually costs $5 for every productive hour, because the other 60% is dead weight on the bill. That makes utilization, not the headline hourly rate, the number to attack: batching jobs, autoscaling down when idle, and packing more work per GPU move the effective cost far more than shopping for a slightly cheaper instance.

Worked example

A $2/hour GPU runs always-on (730 hours) for $1,460/month. At 40% utilization it does 292 productive hours, so the effective cost is $1,460 ÷ 292 = $5/productive hour. If each productive hour earns $6, revenue is $1,752, profit is $292, and ROI is 292 ÷ 1,460 = 20%.

Frequently asked questions

Why is my effective cost higher than the hourly rate?

Because you pay for every hour but only some are productive. Dividing the full bill by just the productive hours reveals the real cost of useful work — the lower the utilization, the bigger the gap.

How do I raise utilization?

Batch and queue jobs so the GPU rarely sits idle, autoscale to zero (or use on-demand spin-up) when there is no work, and co-locate smaller models on one GPU. Reserved or spot pricing lowers the rate but does nothing for idle waste.

What counts as revenue per productive hour?

The value each hour of real GPU work generates — inference requests served and billed, training that ships a product feature, or an internal cost it displaces. For internal workloads, use the price of the cheapest alternative that would do the same job.

Related calculators

Model Training Cost Calculator (GPU-Hours)

Turn a training run — number of GPUs, wall-clock hours and price per GPU-hour — into total GPU-hours and dollar cost, the way cloud accelerator bills are actually rated.

Inference Latency Budget Calculator

Break an LLM response time into time-to-first-token plus generation time, and check it against your target latency budget.

GPU VRAM for Model Calculator

Estimate the GPU memory needed to load an AI model from its parameter count and numeric precision.

Batch Size Memory Calculator

Estimate the activation memory a training batch consumes from batch size, sequence length, hidden size and layer count — the term that scales with batch and blows up GPU memory.

Educational estimate — real costs and memory depend on your provider, framework, model architecture and pricing, which change often. Verify against your own bills and profiling before committing budget.