H100 vs A100
Lower starting price: A100 ($1.16/GPU-hour difference)
Collected on 2026-10-19
Per GPU-hour (VAT excluded)
Base exchange rate: 1 USD = 1,331 KRW (as of 2026-10-20)
Collected on 2026-10-19View collection & calculation basis →
| Feature | NVIDIA H100 | NVIDIA A100 |
|---|---|---|
| Lowest Price | $1.47 | $0.31 Lower |
| Price Range | $1.47–$12.29 | $0.31–$5.12 |
| Providers | 12 providers | 13 providers |
| VRAM | 80GB | 40, 80GB |
| Form Factor | SXM, PCIe | SXM, PCIe |
| Use Cases | Training, Inference | Training, Inference |
NVIDIA H100
| Provider | VRAM | Form Factor | Region | Price / GPU-hr | vs Last Week | Link |
|---|---|---|---|---|---|---|
| Vast.ai | 80GB | SXM | Global | $1.47 Lowest | Same | |
| CoreWeave | 80GB | SXM | US | $2.44 | Same | |
| JarvisLabs | 80GB | SXM | Global | $2.69 | Same | |
| RunPod | 80GB | PCIe | Global | $2.89 | Same | |
| Lambda | 80GB | PCIe | US | $3.29 | Same | |
| RunPod | 80GB | SXM | Global | $3.49 | Same | |
| Elice Cloud | 80GB | SXM | Korea | $3.72₩4,950 | Same | |
| Nebius AI | 80GB | SXM | Global | $3.85 | Same | |
| Elice Cloud | 80GB | SXM | Korea | $4.13₩5,500 | +11.1% | |
| Lambda | 80GB | SXM | US | $4.29 | Same |
NVIDIA A100
| Provider | VRAM | Form Factor | Region | Price / GPU-hr | vs Last Week | Link |
|---|---|---|---|---|---|---|
| Vast.ai | 80GB | SXM | Global | $0.31 Lowest | Same | |
| JarvisLabs | 80GB | SXM | Global | $1.49 | Same | |
| RunPod | 80GB | PCIe | Global | $1.59 | Same | |
| RunPod | 80GB | SXM | Global | $1.59 | Same | |
| Elice Cloud | 80GB | PCIe | Korea | $1.69₩2,250 | Same | |
| Elice Cloud | 80GB | PCIe | Korea | $1.88₩2,500 | +11.1% | |
| Lambda | 40GB | PCIe | US | $1.99 | Same | |
| CoreWeave | 80GB | SXM | US | $2.70 | Same | |
| Lambda | 80GB | SXM | US | $2.79 | Same | |
| Microsoft Azure | 80GB | PCIe | US | $3.67 | Same |
Sign-up is not open yet.
Model Overview
NVIDIA H100
A data center GPU based on the NVIDIA Hopper architecture. Clouds mainly offer it with 80GB of VRAM in SXM or PCIe form. It is used for large-model training and inference.
NVIDIA A100
A data center GPU based on the NVIDIA Ampere architecture. It comes with 40GB or 80GB of VRAM. It is used for both training and inference.
Related Guides
- 70B LLM Serving: Minimum GPU Setup & Hourly Cloud Cost2026-10-20
- Distributed Training GPU Clusters: SXM vs PCIe NVLink Costs2026-10-20
- Elice Cloud GPU Pricing: On-Demand ECI Instance Guide2026-10-20
- Best GPUs for AI Model Fine-Tuning: LoRA vs Full Tuning2026-10-20
- Best Cloud GPUs for LLM Inference: VRAM & Cost Guide2026-10-20
- H100 vs A100 Cloud Cost: Which GPU to Rent?2026-10-20
- Lambda Labs GPU Cloud Pricing: Hourly Rates & Availability2026-10-20
- RunPod Pricing Guide: Secure vs Community Cloud Hourly Rates2026-10-20