Available in synchronized catalog
H200 SXM
Asia-Pacific / Secure
- GPUs per machine
- 1
- VRAM per GPU
- 141 GB
- vCPU
- 12
- System RAM
- 188 GB
- Storage
- 100 GB
USD 5.82 / machine-hour
NVIDIA H200 GPU rental and cloud pricing
Evaluate NVIDIA H200 for memory-intensive model training, inference, and GPU-accelerated AI workloads. Compare the exact configurations below using synchronized customer prices, not a fixed promotional rate.
Latest successful catalog synchronization: . This is a synchronized snapshot, not a reservation or a guarantee of capacity at launch. Prices and availability may change.
Catalog VRAM per GPU: 141 / 143 GB.
From USD 5.82 / machine-hour (1 GPU)
Available NVIDIA H200 configurations (3)
Available in synchronized catalog
H200 SXM
EU West / Secure
- GPUs per machine
- 1
- VRAM per GPU
- 141 GB
- vCPU
- 12
- System RAM
- 188 GB
- Storage
- 100 GB
USD 5.82 / machine-hour
Available in synchronized catalog
H200 SXM
US East / Secure
- GPUs per machine
- 1
- VRAM per GPU
- 141 GB
- vCPU
- 12
- System RAM
- 188 GB
- Storage
- 100 GB
USD 5.82 / machine-hour
When to evaluate NVIDIA H200
Compare the listed NVL, PCIe, or SXM configuration rather than assuming identical memory or connectivity. Validate framework, precision, and multi-GPU requirements before scaling.
For memory-intensive model training, inference, and GPU-accelerated AI workloads, test a representative workload before committing to a larger deployment. Check model weights, runtime allocations, context or scene size, and any application licensing requirements.
Read the configuration and hourly rate together
The displayed rate covers one machine with the listed GPU count. VRAM is per GPU, not system RAM or automatically pooled memory. Prices are customer-facing catalog rates; account-specific terms and configuration changes must be reviewed in the Marketplace before launch.
The public feed chooses the lowest USD customer rate within each model, GPU count, VRAM, region, and capacity class. CPU, RAM, and storage describe that selected listing, not every machine in the region. The number of rows is not a count of available machines.
Read the GPU cloud guide and SSH guide to prepare the software environment and access settings. Capacity shown here is not reserved by viewing the page.