NVIDIA·Datacenter
H100
NVIDIA Hopper architecture GPU designed for generative AI and large language models at data center scale.
Cloud rental pricing
Estimated / month
$1,072
Hourly
$1.47/hr
Data source
Vast.ai
- MSRP
- $25,000
- Availability
- limited
- Providers
- 1
- Reliability
- 1%
Last price update 6/18/2026, 2:20:43 PM
Hardware specifications
- Architecture
- Hopper
- Release
- March 2022
- Memory
- 80 GB
- Memory type
- HBM3
- CUDA cores
- 16,896
- TDP
- 700W
Features
- Transformer Engine
- NVLink
- FP8 precision
- Confidential computing
Use cases
- LLM training
- Generative AI
- Recommendation systems
- Real-time inference
Strengths
- • Fastest NVIDIA datacenter GPU
- • Optimized for transformers
- • High throughput
Considerations
- • Very expensive
- • Limited availability
- • High power and cooling requirements
Typical cloud instance specs
Averages from marketplace listings when rental data is available.
- CPU cores
- 32
- System RAM
- 221115 GB
- Disk
- 492 GB
- Download
- 9374 Mbps
- Upload
- 9227 Mbps
- CUDA
- 12.8
- DL perf
- 358.2
- CPU
- AMD EPYC 9554 64-Core Processor
Price summary
- Low
- $1,072
- Average
- $1,072
- High
- $1,072
Price history
| Date | Price / mo | Source |
|---|---|---|
| 6/18/2026, 2:20:43 PM | $1,072 | Vast.ai |