All GPUs
NVIDIA · Hopper
NVIDIA H100 SXM
The proven workhorse for distributed training and production inference.
- On-demand
- 2 years
- 3 years
Specifications
- Architecture
- Hopper
- GPU memory
- 80 GB HBM3
- Memory bandwidth
- 3.35 TB/s
- GPU-to-GPU
- NVLink 4 · 900 GB/s
- Form factor
- SXM (HGX)
Best for
- Distributed training across NVLink-connected GPUs
- Production LLM inference
- Fine-tuning mid-size and large models