GPU Cost Calculator
Compare cloud GPU costs across 9 providers. See compute costs, electricity expenses, and carbon footprint — find the cheapest option for your workload.
Cheapest Provider
TensorDock
$17.8K total
Max Savings
$24.5K
58% vs most expensive
Electricity Cost
$368
4.2 kW draw
CO₂ Footprint
1.2t
US avg grid (385g/kWh)
Average GPU utilization is 5%. At that rate, 8x H100 SXM wastes $265.6K/year on idle compute.
NemulAI monitors utilization per-job and recommends power caps to eliminate waste. Try free →
Provider Comparison — 8x H100 SXM for 1 month
| Provider | $/hr per GPU | Compute Cost | Electricity | Total | Savings | |
|---|---|---|---|---|---|---|
| TensorDockCHEAPEST | $2.99 | $17.5K | $368 | $17.8K | $24.5K (58%) | |
| FluidStack | $3.19 | $18.6K | $368 | $19.0K | $23.3K (55%) | |
| Vast.ai | $3.39 | $19.8K | $368 | $20.2K | $22.1K (52%) | |
| RunPod | $3.99 | $23.3K | $368 | $23.7K | $18.6K (44%) | |
| Lambda | $4.59 | $26.8K | $368 | $27.2K | $15.1K (36%) | |
| CoreWeave | $4.79 | $28.0K | $368 | $28.3K | $14.0K (33%) | |
| GCP | $6.78 | $39.6K | $368 | $40.0K | $2330 (6%) | |
| Azure | $6.98 | $40.8K | $368 | $41.1K | $1165 (3%) | |
| AWS | $7.18 | $41.9K | $368 | $42.3K | — |
TDP
700W
FP16 Performance
989 TFLOPS
Memory
80 GB
Architecture
Hopper
Stop guessing. Start measuring.
Install the NemulAI agent to get real power measurements, per-job cost attribution, and optimization recommendations for your actual workloads.
pip install nemulaiHow GPU Cloud Pricing Works
Cloud GPU pricing varies dramatically across providers. The same NVIDIA H100 SXM can cost anywhere from $2.99/hr on TensorDock to $7.18/hr on AWS — a 140% difference. This calculator uses baseline rates from 9 major providers including RunPod, Vast.ai, Lambda, CoreWeave, FluidStack, TensorDock, GCP, Azure, and AWS.
Beyond compute costs, electricity is a hidden expense that most calculators ignore. A single H100 running at 75% utilization draws 525W, costing $0.063/hr at the US average rate of $0.12/kWh. For an 8-GPU cluster running 24/7, that adds $4,400/year in electricity alone.
GPU Utilization: The 95% Waste Problem
Research shows average GPU utilization across data centers is just 5%. That means 95% of GPU compute time is wasted — GPUs sitting idle while still drawing significant power. An idle H100 still consumes ~100W. NemulAI's monitoring agent measures actual utilization per job, per GPU, so you can identify and eliminate waste.
EU AI Act Energy Reporting
The EU AI Act (Article 13) requires GPAI model providers to report energy consumption during training. NemulAI provides automated energy metering, carbon footprint calculation, and compliance-ready reports for EU AI Act Article 13 obligations.