GPU comparison
MI325X vs H200
MI325X carries 256 GB against 141 GB, more bandwidth (6 vs 4.8 TB/s) and more dense FP8 (2.6 vs 1.98 PF). H200 is the safer pick for CUDA-only stacks; MI325X fits much larger models per GPU on ROCm.
- Updated Nov 4, 2026
- Specs: vendor datasheets, dense Tensor figures
- Prices: live CoreGPU catalog
| Spec | AMD Instinct MI325X | NVIDIA H200 SXM |
|---|---|---|
| Architecture | CDNA 3 | Hopper |
| GPU memory | 256 GB HBM3E | 141 GB HBM3e |
| Memory bandwidth | 6 TB/s | 4.8 TB/s |
| FP8 dense | 2.6 PF | 1.98 PF |
| FP4 dense | n/a | n/a |
| Interconnect | Infinity Fabric · 896 GB/s | NVLink 4 · 900 GB/s |
| Power (TDP) | 1,000 W | 700 W |
| Price per GPU / month | $2,299 | $2,699 |
| Price per GB of memory | $8.98 | $19.14 |
| Price per FP8 PFLOPS | $884 | $1,363 |
| Availability | Back Dec 2 | 34 available |
Green marks the better value on each line. Prices are per GPU per month at CoreGPU, at least 25% below the market median.
- 400B+ models on fewer GPUs
- Long-context serving on ROCm
- Large-batch fine-tuning
$2,299 /GPU/moJoin waitlist MI325X
- CUDA-only kernels and TensorRT-LLM
- Fine-tuning 70B models with standard recipes
- Lower risk for teams new to ROCm
$2,699 /GPU/moConfigure H200