7/7
Providers w/ Spot
$1.99
Lowest Spot /hr
98%
Max Savings vs On-Demand
Hardware Specifications
Hopper
80 GB HBM3
VRAM
989 TFLOPS
FP16 Performance
67 TFLOPS
FP32 Performance
700W
Thermal Design Power
3.35 TB/s
Memory Bandwidth
Hopper
GPU Architecture
Provider Spot Price On-Demand Savings Availability
Amazon Web Services $35.72/hr $98.32/hr 64% Available
Google Cloud Platform $26.86/hr $88.49/hr 70% Available
Microsoft Azure $4.27/hr $13.96/hr 69% Available
Lambda Cloud $2.49/hr $2.49/hr 0% Available
CoreWeave $2.06/hr $4.76/hr 57% Available
RunPod $1.99/hr $2.89/hr 31% Available
Vast.ai $4.40/hr $6.16/hr 29% Available
What is GPU Spot Pricing?

GPU spot instances use unused cloud capacity at discounts of 60–90% compared to on-demand pricing. The tradeoff: cloud providers can reclaim instances with little notice (typically 30 seconds warning). For fault-tolerant workloads like batch inference, training, and CI/CD, spot pricing dramatically reduces GPU costs.

RoofRun monitors prices from 7 providers every 30 minutes and surfaces the cheapest H100 prices so you don't have to check each provider manually.