H100
80GB SXM5 or 94GB NVL per GPU
View GPU All GPUsH200
Hopper architecture141GB HBM3e per GPU
GPU Catalog / NVIDIA
H200
GPU infrastructure.
A memory-focused Hopper GPU family for large-model inference and data-intensive scientific computing.
- Architecture
- Hopper
- GPU memory
- 141GB HBM3e per GPU
- Published from
- $3.50 / hr
Published pricing
Available configurations.
Current Geodd hourly rates. GPU memory totals reflect the accelerator count in each published configuration.
| Configuration | Variant | Total GPU memory | Hourly rate | Action |
|---|---|---|---|---|
1x GPUgpu_1x_h200_nvl | H200 NVL | 141GB HBM3e141GB HBM3e each | $3.50per hour | Deploy |
2x GPUgpu_2x_h200_nvl | H200 NVL | 282GB HBM3e141GB HBM3e each | $7.00per hour | Deploy |
4x GPUgpu_4x_h200_nvl | H200 NVL | 564GB HBM3e141GB HBM3e each | $14.00per hour | Deploy |
8x GPUgpu_8x_h200_nvl | H200 NVL | 1128GB HBM3e141GB HBM3e each | $28.00per hour | Deploy |
8x GPUgpu_8x_h200_SXM5 | H200 SXM | 1128GB HBM3e141GB HBM3e each | $29.00per hour | Deploy |
Hardware profile
Where H200 fits.
H200 extends the Hopper generation with 141GB of HBM3e memory per GPU. Geodd pricing can include NVL and SXM configurations, with each variant identified separately in the configuration table.
Choose H200 when memory capacity and bandwidth are central constraints, especially for large language model inference or scientific workloads with large working sets.
NVIDIA product information- 01
Large-model inference
Use the 141GB HBM3e capacity for language-model serving where model weights and runtime state place sustained pressure on GPU memory.
- 02
Memory-intensive HPC
Run scientific simulations and high-performance computing applications that benefit from Hopper compute and a larger HBM pool.
- 03
Multi-GPU capacity
Scale total GPU memory through the published multi-GPU configurations when one accelerator is not enough for the workload.
Deployment sizing
Size before you provision.
Hardware fit depends on the model, precision, batch size, software stack, and total working set. Published use cases are not performance guarantees.
- 01 / Memory
Confirm the working set
Include weights, caches, activations, batches, and framework overhead.
- 02 / Software
Validate the stack
Check CUDA, driver, framework, precision, and container requirements.
- 03 / Capacity
Confirm provisioning
Published pricing does not guarantee immediate capacity in every configuration.