GPU Rental Cost Calculator
Estimate GPU rental prices for AI training, inference, rendering, and notebook workloads. Model hourly rate, GPU count, utilization, storage, network, support overhead, and buy-vs-rent break-even.
Built from 14 providers and 60 GPU families where pricing is available. Last pricing refresh: Jan 15, 2026.
Calculate GPU Rental Cost
Estimated GPU rental price
GPU rental price index
Use this index to choose a realistic starting rate before modeling utilization. The lowest hourly price is useful for screening, while the median tracked price is often safer for budgets that need capacity, region choice, or better reliability.
| GPU | VRAM | Lowest on-demand | Median on-demand | Lowest spot | Providers | Use in calculator |
|---|---|---|---|---|---|---|
| RTX A4000 | 16GB | $0.060 | $0.320 | $0.160 | 4 | Estimate cost |
| RTX 3090 | 24GB | $0.081 | $0.200 | $0.062 | 3 | Estimate cost |
| RTX 3070 | 8GB | $0.130 | $0.130 | $0.070 | 1 | Estimate cost |
| V100 | 32GB | $0.140 | $0.800 | $0.992 | 5 | Estimate cost |
| NVIDIA GeForce RTX 3080 | 10GB | $0.170 | $0.170 | $0.090 | 1 | Estimate cost |
| NVIDIA GeForce RTX 3080 Ti | 12GB | $0.180 | $0.180 | $0.090 | 1 | Estimate cost |
| V100 FHHL | 16GB | $0.190 | $0.190 | $0.100 | 1 | Estimate cost |
| Tesla V100 | 16GB | $0.190 | $0.190 | $0.100 | 1 | Estimate cost |
| V100 SXM2 | 16GB | $0.230 | $0.230 | $0.120 | 1 | Estimate cost |
| RTX 4090 | 24GB | $0.235 | $0.350 | $0.109 | 3 | Estimate cost |
| RTX 2000 Ada | 16GB | $0.240 | $0.240 | $0.140 | 1 | Estimate cost |
| RTX A5000 | 24GB | $0.250 | $0.490 | $0.140 | 5 | Estimate cost |
| RTX A4500 | 20GB | $0.250 | $0.250 | $0.180 | 1 | Estimate cost |
| RTX 4000 Ada | 20GB | $0.260 | $0.260 | $0.190 | 1 | Estimate cost |
| A100 PCIe | 80GB | $0.280 | $0.280 | $0.160 | 1 | Estimate cost |
| T4 | 16GB | $0.350 | $0.978 | $0.140 | 3 | Estimate cost |
| L4 | 24GB | $0.390 | $2.81 | $0.220 | 3 | Estimate cost |
| A40 | 48GB | $0.400 | $0.800 | $0.200 | 3 | Estimate cost |
Example GPU rental cost scenarios
| Scenario | GPU count | Schedule | Rate | Utilization | Estimated compute cost |
|---|---|---|---|---|---|
| Short AI notebook test | 1 | 6 hr/day x 3 days | $0.450/GPU-hr | 60% | $5.35 |
| Fine-tuning sprint | 4 | 10 hr/day x 14 days | $1.80/GPU-hr | 75% | $869.40 |
| Always-on inference pilot | 2 | 24 hr/day x 30 days | $1.25/GPU-hr | 85% | $1.7k |
| Training cluster month | 32 | 18 hr/day x 30 days | $2.99/GPU-hr | 70% | $43.4k |
What changes the real GPU rental price?
Listed rate vs completed-job cost
A cheap hourly rate can be expensive if the job runs longer, fails often, or waits on slow storage and data loading.
Spot, interruptible, and reserved pricing
Spot works for checkpointed jobs. Reserved or committed capacity can beat on-demand when usage is predictable for months.
GPU memory fit
A lower-priced GPU is not cheaper if the model spills to CPU, requires more replicas, or needs tensor parallelism to fit in memory.
Provider type
Marketplaces can be cheaper but vary by host reliability. Hyperscalers cost more but may include stronger networking, IAM, compliance, and support.
Storage and network fees
Persistent volumes, egress, object storage, local NVMe, and cross-region transfer can move the final bill beyond headline GPU rental prices.
Idle time and orchestration
Stopped notebooks, warm endpoints, Kubernetes overhead, logs, and failed experiments should be included in practical GPU rental budgets.
Lowest tracked provider rates by GPU
| GPU | Provider | Instance | On-demand / GPU-hour | Spot / GPU-hour | Provider type |
|---|---|---|---|---|---|
| RTX A4000 | TensorDock | tensordock-rtx-a4000 | $0.060 | - | marketplace |
| RTX 3090 | Vast.ai | vastai-rtx-3090 | $0.081 | $0.062 | marketplace |
| RTX 3070 | RunPod | NVIDIA GeForce RTX 3070 | $0.130 | $0.070 | gpu-cloud |
| V100 | Datacrunch (Verda) | datacrunch-v100-32gb | $0.140 | - | gpu-cloud |
| NVIDIA GeForce RTX 3080 | RunPod | NVIDIA GeForce RTX 3080 | $0.170 | $0.090 | gpu-cloud |
| NVIDIA GeForce RTX 3080 Ti | RunPod | NVIDIA GeForce RTX 3080 Ti | $0.180 | $0.090 | gpu-cloud |
| V100 FHHL | RunPod | Tesla V100-FHHL-16GB | $0.190 | $0.100 | gpu-cloud |
| Tesla V100 | RunPod | Tesla V100-PCIE-16GB | $0.190 | $0.100 | gpu-cloud |
| V100 SXM2 | RunPod | Tesla V100-SXM2-16GB | $0.230 | $0.120 | gpu-cloud |
| RTX 4090 | Vast.ai | vastai-rtx-4090 | $0.235 | $0.109 | marketplace |
| RTX 2000 Ada | RunPod | NVIDIA RTX 2000 Ada Generation | $0.240 | $0.140 | gpu-cloud |
| RTX A5000 | Genesis Cloud | genesis-rtx-a5000 | $0.250 | - | gpu-cloud |
| RTX A4500 | RunPod | NVIDIA RTX A4500 | $0.250 | $0.180 | gpu-cloud |
| RTX 4000 Ada | RunPod | NVIDIA RTX 4000 Ada Generation | $0.260 | $0.190 | gpu-cloud |
For full filters, provider pages, and all available rows, use the cloud GPU pricing comparison.
When to use this calculator instead of a single-GPU page
Use this page when
- You are comparing GPU rental prices across H100, A100, L40S, RTX 4090, MI300X, B200, and other models.
- You need a project budget before choosing a provider or committing to reserved capacity.
- You want to include storage, network, support, idle time, and failure overhead in the same estimate.
- You need a quick buy-vs-rent threshold for a cluster or repeated workload.
Use a focused calculator when
- H100 rental cost is the main decision and you need H100-specific capacity notes.
- A100 rental cost matters because the workload fits mature A100 capacity.
- H200 rental cost is relevant for large-memory HBM3e workloads.
- RTX 5090 rental cost is enough for ComfyUI, rendering, or consumer-GPU experiments.
Useful GPU rental pricing references
GPU Cost tracks normalized rows where available, but final budgets should check current provider pages and public market trackers: Vast.ai live GPU pricing, Runpod GPU cloud pricing, Salad GPU pricing, AIMultiple GPU rental price index, and SemiAnalysis GPU Pricing Index.