NVIDIA H100 SXM
Hopper Architecture
Active
Launched September 2022
Core Specifications
VendorNVIDIA
ArchitectureHopper
Form FactorSXM
VRAM80 GB
Memory Bandwidth3,350 GB/s
TDP700 W
Compute Performance
| Precision | TFLOPs |
|---|---|
| FP32 | 67 |
| FP16 | 1,979 |
| BF16 | 1,979 |
| FP8 | 3,958 |
Performance Benchmarks
image gen
| Configuration | Precision | Performance | Source |
|---|---|---|---|
| Stable Diffusion XL, 1024x1024, 50 steps | — | 2.375 images_per_second | View |
llm inference
| Configuration | Precision | Performance | Source |
|---|---|---|---|
| LLaMA 70B, batch_size=1 | — | 7,600 tokens_per_second | View |
Pricing
Hardware Purchase (CAPEX)
| Type | Price (USD) | Region | As of |
|---|---|---|---|
| Street Price | $32,000 | Global | Oct 2024 |
| Street Price | $30,000 | global | Dec 2025 |
Cloud Rental (OPEX)
| Provider | Instance Type | Price per Hour | Region | As of |
|---|---|---|---|---|
| Salad | 1x H100 (94GB NVL) | $0.99/hr | Global | Oct 2024 |
| Azure | Standard_ND96isr_H100_v5 (8x H100) | $119.60/hr | East US | Oct 2024 |
| Azure | Standard_ND96isr_H100_v5 (1x H100) | $14.95/hr | East US | Oct 2024 |
| Google Cloud | a3-highgpu-8g (8x H100) | $117.84/hr | us-central1 | Oct 2024 |
| Google Cloud | a3-highgpu-1g (1x H100) | $14.73/hr | us-central1 | Oct 2024 |
| AWS | p5.48xlarge (8x H100) | $98.32/hr | us-east-1 | Oct 2024 |
| AWS | p5.xlarge (1x H100) | $12.29/hr | us-east-1 | Oct 2024 |
| Lambda Labs | — | $3.50/hr | us-east | Dec 2025 |