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.

$1.38starter / GPU-hour
$995.4724/7 monthly baseline
14tracked providers

Built from 14 providers and 60 GPU families where pricing is available. Last pricing refresh: Jan 15, 2026.

Calculate GPU Rental Cost

$
$
total = hourly_rate x GPU_count x hours_per_day x days x utilization + overhead + storage/network/support

Estimated GPU rental price

Total project cost
$0
Monthly 24/7 equivalent
$0
Effective GPU-hours
0
Buy break-even
0 hours

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

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.

GPU Rental Cost FAQ

GPU rental prices range from low-cost consumer GPUs under a dollar per hour to H100, H200, B200, and large multi-GPU nodes that can cost several dollars per GPU-hour or much more per full instance. The calculator lets you model the actual schedule and overhead instead of relying only on the list rate.

For early screening, use dollars per GPU-hour. For real budgets, compare cost per completed job, cost per million tokens, cost per image, or monthly cost at your expected utilization. The cheapest hourly GPU is not always the cheapest finished workload.

Spot or interruptible GPUs can be useful for checkpointed experiments, batch inference, and resumable jobs. They are risky for long training runs without checkpointing because interruptions, retries, and queue delays can erase the hourly discount.

Buying starts to make sense when usage is steady for months, the workload can run locally or in your own rack, and you can manage power, cooling, networking, maintenance, and resale risk. Short projects, uncertain demand, and bursty experiments usually favor renting.

Choose by memory fit, throughput per dollar, framework support, and reliability needs. H100 and B200 suit high-end training and inference, A100 is mature and widely available, L40S is strong for inference and rendering, and RTX 4090 or RTX 5090 can be cost-effective when consumer GPU memory is enough.