Side by side
| H100 | H200 | |
|---|---|---|
| GPU memory | 80 GB HBM3 | 141 GB HBM3e |
| Memory bandwidth | 3.35 TB/s | 4.8 TB/s |
| BF16 dense | ≈ 990 TFLOPS | ≈ 990 TFLOPS |
| NVLink | 900 GB/s | 900 GB/s |
| Strata price | from $3.23/GPU-hr | from $4.31/GPU-hr |
Which one to choose
Choose H100 for training and for models that fit comfortably in 80 GB: it is cheaper per hour and the most widely supported GPU.
Choose H200 for large-model inference, long contexts and high concurrency: more and faster memory means fewer GPUs per replica and more tokens per second, which often makes it cheaper per token.




