H200 SXM
LLM inference · fine-tuning
$1,832/ 30 days
View GPU H200 SXM vs A40
Compare memory, node configurations, and fixed weekly and monthly cost before choosing between NVIDIA H200 SXM and NVIDIA A40.
H200 SXM provides 93 GB more VRAM per card. A40 costs $1,640 less per 30-day term . Choose based on whether memory headroom or lower fixed cost is the binding constraint.
LLM inference · fine-tuning
rendering · inference
| Specification | H200 SXM | A40 |
|---|---|---|
| Architecture | Hopper | Ampere |
| VRAM | 141 GB HBM3e | 48 GB GDDR6 ECC |
| Host resources / card | 24 vCPU · 192 GB RAM | 12 vCPU · 96 GB RAM |
| Local NVMe / card | 3.84 TB | 1 TB |
| 7-day price / card | $512 | $54 |
| 30-day price / card | $1,832 | $192 |
| 4-GPU node / 30 days | $6,742 | $707 |
| Catalog quantity | 265 cards | 265 cards |